{"ok":true,"key":"sentiment-bench","name":"Sentiment Bench","about":"How people received the model in its first week on X, Hacker News, in the news and among trusted reviewers (Handy AI).","type":{"key":"sentiment","label":"Sentiment","folded":true},"unit":null,"lower_is_better":false,"weight_cap":"median","url":"https://handyai.news/benchmarks/sentiment-bench","api_url":"https://handyai.news/api/benchmarks/sentiment-bench","image_url":"https://handyai.news/api/benchmarks/sentiment-bench.png","leader":{"model":"Claude Opus 4.5","score":91.5,"maker":"Anthropic"},"model_count":83,"drop_count":79,"first_date":"2025-10-27","last_date":"2026-10-28","frontier_gain":14.2,"frontier":[{"date":"2025-10-27","model":"MiniMax-M2","score":77.3,"maker":"minimax"},{"date":"2025-10-29","model":"Composer","score":84.1,"maker":"cursor"},{"date":"2025-11-24","model":"Claude Opus 4.5","score":91.5,"maker":"anthropic"}],"models":[{"model":"Claude Opus 4.5","score":91.5,"own_drop":true,"independent":false,"setting":"","source":{"drop":"Claude Opus 4.5","slug":"claude-opus-45","url":"https://handyai.news/benchmarks/sentiment-bench#claude-opus-45","published":"2025-11-24"},"launch_score":91.5,"reports":1,"maker":{"key":"anthropic","name":"Anthropic","logo_url":"https://handyai.news/ui/makers/anthropic.svg"},"open_weights":false,"released":"2025-11-24","dated_by":"own_drop","drop_url":"https://handyai.news/modeldrop/claude-opus-45","rank":1,"sentiment":{"reception":94.2,"lives_up_to_benchmarks":83.3,"opinions":129,"hands_on":49,"crowd":91.7,"sources":{"x":{"score":93.5,"adjusted":93.8,"typical":76.9,"opinions":25,"positive":23,"negative":2},"hn":{"score":63.2,"adjusted":91.6,"typical":48.4,"opinions":79,"positive":51,"negative":30},"news":{"score":97.1,"adjusted":87.3,"typical":89.1,"opinions":21,"positive":21,"negative":0}},"reviewers":{"score":91.5,"adjusted":100,"count":4,"each":[{"name":"Every","score":100,"url":"https://every.to/vibe-check/vibe-check-opus-4-5-is-the-coding-model-we-ve-been-waiting-for"},{"name":"Ethan Mollick","score":100,"url":"https://x.com/emollick/status/1993030988759470156"},{"name":"Simon Willison","score":100,"url":"https://simonwillison.net/2025/Nov/24/claude-opus/"},{"name":"Zvi Mowshowitz","score":100,"url":"https://thezvi.substack.com/p/claude-opus-45-model-card-alignment"}]},"verdicts":{"beats":0,"matches":2,"falls_short":6},"buzz":{"x_posts_that_week":null,"hn_points":1138,"articles":47},"quotes":{"positive":[{"source":"x","text":"been using opus 4.5 for the past week and it changes everything i think we're reaching a new heights on what is possible with coding agents i do not use my editor anymore something something step function jk we don't work with model labs so we don't have to say this","url":"https://x.com/thdxr/status/1993057593213038605","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"To my surprise, Opus 4.5 one-shot my hardest λ-calculus problem (tying with Gemini 3), and it did solve the stack underflow bug that an old checkpoint of Gemini 3 (NOT the deployed version) solved. So, in terms of first hour impression, that couldn't be more promising I guess...","url":"https://x.com/VictorTaelin/status/1993042103048540619","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"The new @AnthropicAI Opus 4.5 model is absolutely stunningly good at literary analysis. I'm helping a friend with their book and there's times I feel like something is a little off in the writing, and when I ask Opus about it, it always diagnoses it beautifully, with examples.","url":"https://x.com/jeremyphoward/status/1995200930384650697","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"}],"negative":[{"source":"x","text":"I'm not enjoying Claude with Opus 4.5. It no longer tells me “I'm absolutely right”. My code worked on the first try today but at what cost. pic.twitter.com/zKPolPeUQs","url":"https://x.com/deedydas/status/1993376627452260738","author":null,"stance":-1,"hands_on":true,"basis":"used","verdict":"none"},{"source":"hn","text":"I was excited to see this today, as I just finished a new website in Figma last night and planned to have CC help build it out with the Figma MCP today. CC started up with Opus 4.5 and... Not off to a good start. It described a totally different looking layout on the first prompt","url":"https://news.ycombinator.com/item?id=46039879","author":"pwython","stance":-2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"My Claude 4.5 Opus experience so far 👇 1. I bought the $20 premium plan 2. Asked it to build a 3D room decorator with Three.js 3. Hit the context limit. Pressed continue. Hit it again. And again. 4. I ran into the daily message limit So in the end I never actually tried… pic.twi","url":"https://x.com/flavioAd/status/1993053769488105888","author":null,"stance":-1,"hands_on":true,"basis":"used","verdict":"none"}]},"window":{"start":null,"end":null}}},{"model":"GLM 5.2","score":88,"own_drop":true,"independent":false,"setting":"","source":{"drop":"GLM 5.2","slug":"glm-52","url":"https://handyai.news/benchmarks/sentiment-bench#glm-52","published":"2026-06-19"},"launch_score":88,"reports":1,"maker":{"key":"zai","name":"Z.ai","logo_url":"https://handyai.news/ui/makers/zai.svg"},"open_weights":true,"released":"2026-06-20","dated_by":"own_drop","drop_url":"https://handyai.news/modeldrop/glm-52","rank":2,"sentiment":{"reception":91.2,"lives_up_to_benchmarks":78.5,"opinions":180,"hands_on":54,"crowd":92.2,"sources":{"x":{"score":89.9,"adjusted":89.3,"typical":76.9,"opinions":43,"positive":37,"negative":6},"hn":{"score":68.4,"adjusted":98.1,"typical":48.4,"opinions":113,"positive":75,"negative":38},"news":{"score":97.7,"adjusted":88.3,"typical":89.1,"opinions":21,"positive":21,"negative":0}},"reviewers":{"score":78.5,"adjusted":89,"count":3,"each":[{"name":"Nathan Lambert","score":100,"url":"https://www.interconnects.ai/p/glm-52-is-the-step-change-for-open"},{"name":"Zvi Mowshowitz","score":75,"url":"https://thezvi.substack.com/p/glm-52-is-the-new-best-open-model"},{"name":"Sebastian Raschka","score":75,"url":"https://x.com/rasbt/status/2067612153020838055"}]},"verdicts":{"beats":2,"matches":19,"falls_short":8},"buzz":{"x_posts_that_week":null,"hn_points":1721,"articles":40},"quotes":{"positive":[{"source":"x","text":"Wow. @Zai_org GLM 5.2 is a marvel! It is *at least* as good as Opus 4.8 and GPT 5.5. It's super fast, inexpensive, and not too verbose. It responds with nuance and judgement, & handles long context VERY well. I've never experienced an open weights model like this before.","url":"https://x.com/jeremyphoward/status/2067757468189679764","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"After using GLM-5.2 for a day, I'm surprised by how often it feels close to Opus 4.8/GPT-5.5 level. I compared it side by side with Opus 4.8, and sometimes I even preferred GLM-5.2's results. OSS LLMs are impressive, especially given how many fewer GPUs they were trained on.","url":"https://x.com/Yuchenj_UW/status/2068182756132376668","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"GLM 5.2 is first chinese model that made 3d watch without any major mistakes 🤯 @Zai_org you guys cooked very hard with this model , now only thing that we need is the big fat model , tell us the time when we can expect This is one of the hardest thing to one shot https://t.co/qW","url":"https://x.com/chetaslua/status/2067571640582709468","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"}],"negative":[{"source":"hn","text":"My anecdotal experience differs (though I hold ground that LLM evaluations are highly subjective and benchmarks are just as useful for LLMs as they are for dating websites users). GLM 5.2 tends to stray way more than and 5.1. It also hallucinates you things subtly: morphs require","url":"https://news.ycombinator.com/item?id=48607545","author":"xlii","stance":-2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"Controversial take: if you like GLM 5.2, you might find you're more than happy with GPT 5.4. I know this might sound weird, but as I've been doing more testing with GLM 5.2, I have found that yes, it's a very good model, but it's not the best model, and it definitely has… pic.twi","url":"https://x.com/morganlinton/status/2070151992270549237","author":null,"stance":-1,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"my only complaint about GLM5.2 so far is mode switching into other langs. but then i found out that Anthropic defaults to top_p=0.99 for all API requests nowadays, & doesn't support setting anything else for Opus 4.7 and beyond... which makes tail sampling comparisons unfair! pic","url":"https://x.com/kalomaze/status/2068543918606880841","author":null,"stance":-1,"hands_on":true,"basis":"used","verdict":"none"}]},"window":{"start":null,"end":null}}},{"model":"Grok 4.6","score":87,"own_drop":true,"independent":false,"setting":"","source":{"drop":"Grok 4.6","slug":"grok-46","url":"https://handyai.news/benchmarks/sentiment-bench#grok-46","published":"2026-08-12"},"launch_score":87,"reports":1,"maker":{"key":"xai","name":"SpaceXAI","logo_url":"https://handyai.news/ui/makers/xai.svg"},"open_weights":false,"released":"2026-08-12","dated_by":"own_drop","drop_url":"https://handyai.news/modeldrop/grok-46","rank":3,"sentiment":{"reception":89.2,"lives_up_to_benchmarks":80.2,"opinions":155,"hands_on":57,"crowd":89.2,"sources":{"x":{"score":89.8,"adjusted":89.2,"typical":76.9,"opinions":45,"positive":38,"negative":7},"hn":{"score":61.6,"adjusted":89.6,"typical":48.4,"opinions":88,"positive":53,"negative":46},"news":{"score":97.8,"adjusted":88.5,"typical":89.1,"opinions":22,"positive":22,"negative":0}},"reviewers":{"score":null,"adjusted":null,"count":0,"each":[]},"verdicts":{"beats":0,"matches":2,"falls_short":5},"buzz":{"x_posts_that_week":null,"hn_points":1007,"articles":60},"quotes":{"positive":[{"source":"x","text":"Grok 4.6 worked non-stop for 48 hours to build this shooter. Turns out Grok is powerful enough to run Gauntlet Loops. Let the game-making begin! pic.twitter.com/HbpS1QmEsp","url":"https://x.com/mattshumer_/status/2088303001299529981","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"Happy Anthropic Rug Pull Eve everyone. Enjoy your last day of Claude limits. The +50% weekly boost ends tomorrow, Aug 19. I already cancelled my Claude Max ahead of this. Grok 4.6 is now my daily driver. GPT 5.6 Sol is what I reach for on the hardest tasks. Fable 5 and… pic.twitt","url":"https://x.com/bridgemindai/status/2089694496472678820","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"I switched https://t.co/5u9qesKibT 's auto app and landing builder to @xai Grok now Grok 4.6 is apparently as good as Fable 5 (without the constant preaching and blocking you from doing anything due to SeCuRiTy) https://t.co/Q4ux0yn3Vk pic.twitter.com/h3vgRR4jrd","url":"https://x.com/levelsio/status/2087579763158216795","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"matches"}],"negative":[{"source":"x","text":"I have no idea why but Grok-4.6 just totally freaked out doing some maintenance tasks when it saw my low github id. pic.twitter.com/TGcVPR0Anl","url":"https://x.com/tobi/status/2087704273438052632","author":null,"stance":-1,"hands_on":true,"basis":"used","verdict":"falls_short"},{"source":"x","text":"I tested Gemini 3.7 Flash vs Qwen 3.8 vs Grok 4.6 vs GLM 5.3 Same prompt, 2 scenes: a harvester and farmers working. honestly? Qwen 3.8 impressed me. it just did what I asked. Grok 4.6 failed on coloring. Qwen>GLM> Grok> Gemini which one did better for you? pic.twitter.com/kCN7zG","url":"https://x.com/Bhavani_00007/status/2088646139310645675","author":null,"stance":-1,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"I've been running a lot of tests on Grok 4.6 xhigh in Cursor recently. What I can say is it tends to give much more often than GPT 5.6 Sol Max. The latter was a different beast running for hours, while Grok 4.6 ran 20 minutes max. imo Grok 4.6 is just lazy asf","url":"https://x.com/fahnx8/status/2089287364682145891","author":null,"stance":-2,"hands_on":true,"basis":"used","verdict":"none"}]},"window":{"start":null,"end":null}}},{"model":"Gemini 4 Carbon","score":85.8,"own_drop":true,"independent":false,"setting":"","source":{"drop":"Gemini 4 Carbon","slug":"gemini-4-carbon","url":"https://handyai.news/benchmarks/sentiment-bench#gemini-4-carbon","published":"2026-10-28"},"launch_score":85.8,"reports":1,"maker":{"key":"google","name":"Google DeepMind","logo_url":"https://handyai.news/ui/makers/google.svg"},"open_weights":false,"released":"2026-10-28","dated_by":"own_drop","drop_url":"https://handyai.news/modeldrop/gemini-4-carbon","rank":4,"sentiment":{"reception":85.2,"lives_up_to_benchmarks":87.8,"opinions":98,"hands_on":11,"crowd":81.6,"sources":{"x":{"score":89,"adjusted":88.2,"typical":76.9,"opinions":45,"positive":38,"negative":9},"hn":{"score":47,"adjusted":71.5,"typical":48.4,"opinions":43,"positive":16,"negative":43},"news":{"score":95.5,"adjusted":84.5,"typical":89.1,"opinions":9,"positive":9,"negative":0}},"reviewers":{"score":81.3,"adjusted":93.5,"count":1,"each":[{"name":"Peter Gostev","score":100,"url":"https://x.com/petergostev/status/2115495845894226110"}]},"verdicts":{"beats":4,"matches":2,"falls_short":3},"buzz":{"x_posts_that_week":null,"hn_points":1356,"articles":60},"quotes":{"positive":[{"source":"x","text":"Gemini Carbon 4 is genuinely shocked me < mind you this will be half the price of opus 5.5 > I prompted It for all heroes action sequence according to its lore in comics Avengers Fighting with each other threes , wait for end ironman to doomsday transformation https://t.co/bM9C2Y","url":"https://x.com/chetaslua/status/2116757313575145683","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"beats"},{"source":"x","text":"🚨 Gemini 4 Carbon: JS loop for sleep with original music and svg of girl in rain This model is so hardworking it coded 4000 lines one shot remember google model used to be lazy , i think they fixed that and also carbon have less slop ui compared to Gemini 4 Argon pic.twitter.com","url":"https://x.com/chetaslua/status/2115935747798950069","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"beats"},{"source":"x","text":"Google has reportedly been testing Gemini 4 Carbon on Arena under Argon 4 since October 31st. It seems to be one of the best models among those announced to date. Stunning outputs 👀 https://t.co/oXCDor0089 pic.twitter.com/SPjBsukQkn","url":"https://x.com/testingcatalog/status/2117009417342468398","author":null,"stance":2,"hands_on":false,"basis":"evidence","verdict":"none"}],"negative":[{"source":"x","text":"Carbon is benchmaxxed bullshit, just like Argon. I have already watched youtube videos of what Gemini 4 Carbon does in 3D applications, and visually it is easy to see that it is not as good as four other models, Opus 5.5, Fable 5.1, Sol GPT-6.1, and Astra 6. Carbon's 3D results… ","url":"https://x.com/worksalt/status/2116315447796945015","author":null,"stance":-2,"hands_on":false,"basis":"evidence","verdict":"falls_short"},{"source":"x","text":"Gemini 4 Carbon just became the world's #1 AI model, but its hallucination rate has nearly TRIPLED. Gemini 4 Argon: 4% Gemini 4 Carbon: 11% That's a 2.75x increase in hallucinations despite Carbon being the more intelligent model. pic.twitter.com/2e7dou7gu8","url":"https://x.com/StatsWire/status/2115888901927710738","author":null,"stance":-1,"hands_on":false,"basis":"evidence","verdict":"none"},{"source":"hn","text":"Carbon scoring below Opus 5.5 on coding highlights how magnificent Opus 5.5 is as a coding agent. With just a $20/month sub, it performs a ton of work at a very high quality.","url":"https://news.ycombinator.com/item?id=50288964","author":"x187463","stance":-1,"hands_on":false,"basis":"evidence","verdict":"none"}]},"window":{"start":null,"end":null}}},{"model":"GPT-5.3-Codex","score":85.6,"own_drop":true,"independent":false,"setting":"","source":{"drop":"GPT-5.3-Codex","slug":"gpt-53-codex","url":"https://handyai.news/benchmarks/sentiment-bench#gpt-53-codex","published":"2026-02-05"},"launch_score":85.6,"reports":1,"maker":{"key":"openai","name":"OpenAI","logo_url":"https://handyai.news/ui/makers/openai.svg"},"open_weights":false,"released":"2026-02-05","dated_by":"own_drop","drop_url":"https://handyai.news/modeldrop/gpt-53-codex","rank":5,"sentiment":{"reception":86.7,"lives_up_to_benchmarks":82.1,"opinions":107,"hands_on":41,"crowd":85.7,"sources":{"x":{"score":88.4,"adjusted":87.4,"typical":76.9,"opinions":40,"positive":33,"negative":7},"hn":{"score":58.3,"adjusted":85.5,"typical":48.4,"opinions":55,"positive":31,"negative":25},"news":{"score":94.3,"adjusted":82.4,"typical":89.1,"opinions":8,"positive":8,"negative":0}},"reviewers":{"score":78.5,"adjusted":89,"count":3,"each":[{"name":"Max Woolf","score":100,"url":"https://news.ycombinator.com/item?id=46940469"},{"name":"Simon Willison","score":75,"url":"https://x.com/simonw/status/2019508962241196296"},{"name":"Every","score":75,"url":"https://every.to/vibe-check/codex-vs-opus"}]},"verdicts":{"beats":2,"matches":0,"falls_short":3},"buzz":{"x_posts_that_week":null,"hn_points":1607,"articles":20},"quotes":{"positive":[{"source":"x","text":"Early impressions on Codex 5.3 and Opus 4.6 I spent the day vibe-coding and testing the models Short version: - Both models are significantly better - Codex 5.3 is *way* faster than 5.2, which I enjoy the most - Codex 5.3 produced more correct code than Opus 4.6 - Opus 4.6…","url":"https://x.com/VictorTaelin/status/2019541668517617859","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"GPT-5.3 Codex is actually pretty insane with Three.js This Minecraft clone works smoothly and it didn't take too long to make I also tried Opus 4.6, but for some reason it got stuck pic.twitter.com/y0WdXqtmXi","url":"https://x.com/Angaisb_/status/2019486685763911867","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"GPT-5.3-Codex with image gen skill (NBP) I'm loving this model so much. This is one-shot pic.twitter.com/uJsLKueLJC","url":"https://x.com/Angaisb_/status/2019548783869325331","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"}],"negative":[{"source":"x","text":"so I just attempted to vibe-code an iOS app with Swift and Codex 5.3. I never used Swift before. I just asked GPT to add features without telling it how it should do it, and with no idea how things work it failed miserably I guess I get how most people experience AI now","url":"https://x.com/VictorTaelin/status/2021329900674650302","author":null,"stance":-1,"hands_on":true,"basis":"used","verdict":"none"},{"source":"hn","text":"Today I also discovered that the speed of gpt-5.3-codex in Codex CLI is extremely slow, and then I found that response.model was routed back to gpt-5.2-2025-12-11 by the upstream.","url":"https://news.ycombinator.com/item?id=46968976","author":"ekaesmem","stance":-2,"hands_on":true,"basis":"used","verdict":"falls_short"},{"source":"x","text":"GPT-5.3-codex early review: this thing runs auto compaction like a maniac. it doesn't wait for context to fill up and instead starts garbage collecting tokens the second it decides they're dead weight. also it gives you updates every few steps like \"hey here's what I am…","url":"https://x.com/cto_junior/status/2019607817884475718","author":null,"stance":-1,"hands_on":true,"basis":"used","verdict":"none"}]},"window":{"start":null,"end":null}}},{"model":"Claude Opus 5.5","score":85.2,"own_drop":true,"independent":false,"setting":"","source":{"drop":"Claude Opus 5.5","slug":"claude-opus-55","url":"https://handyai.news/benchmarks/sentiment-bench#claude-opus-55","published":"2026-09-22"},"launch_score":85.2,"reports":1,"maker":{"key":"anthropic","name":"Anthropic","logo_url":"https://handyai.news/ui/makers/anthropic.svg"},"open_weights":false,"released":"2026-09-22","dated_by":"own_drop","drop_url":"https://handyai.news/modeldrop/claude-opus-55","rank":6,"sentiment":{"reception":86.5,"lives_up_to_benchmarks":81.6,"opinions":204,"hands_on":87,"crowd":87.3,"sources":{"x":{"score":92.3,"adjusted":92.3,"typical":76.9,"opinions":45,"positive":41,"negative":4},"hn":{"score":59.6,"adjusted":87.2,"typical":48.4,"opinions":124,"positive":64,"negative":65},"news":{"score":90.7,"adjusted":76,"typical":89.1,"opinions":28,"positive":26,"negative":2}},"reviewers":{"score":75.8,"adjusted":84.6,"count":6,"each":[{"name":"Ethan Mollick","score":100,"url":"https://x.com/emollick/status/2104039050080342456"},{"name":"Every","score":100,"url":"https://every.to/vibe-check/vibe-check-opus-5-5-is-pulling-our-codex-converts-back-to-claude"},{"name":"Zvi Mowshowitz","score":90,"url":"https://thezvi.substack.com/p/claude-opus-55-should-raise-your"},{"name":"Simon Willison","score":75,"url":"https://simonwillison.net/2026/Sep/22/opus-and-sol-and-luna/"},{"name":"Max Woolf","score":75,"url":"https://news.ycombinator.com/item?id=49837044"},{"name":"xlr8harder","score":25,"url":"https://x.com/xlr8harder/status/2102638211549835423"}]},"verdicts":{"beats":6,"matches":11,"falls_short":8},"buzz":{"x_posts_that_week":null,"hn_points":2572,"articles":60},"quotes":{"positive":[{"source":"x","text":"Opus 5.5 is incredible at instructional video generation. I made this launch video for a inference startup in 1min for ~$2. Videos like these used to take weeks if not months and a lot of coordination with agencies and 1000x the costs. Humans broadly prefer video to text. This… p","url":"https://x.com/deedydas/status/2102787937482252537","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"Using Claude regularly again has made obvious one of the biggest differences between Anthropic and OpenAI models: taste GPT (Astra included) just doesn't really compete. Sure, you can get ~similar quality outputs with enough prompting, time and effort, but Opus 5.5 just combines…","url":"https://x.com/synthwavedd/status/2103192588048310404","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"The more I use, the more I fall in love with opus 5.5. its so good and so much faster and so much less verbose. It's everything I could have asked for. It's as if my beloved Opus 4.6 came back but better and rebranded as 5.5","url":"https://x.com/kimmonismus/status/2102515704318636296","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"}],"negative":[{"source":"x","text":"I caved and bought Claude Pro to see what all the rage is about. Opus 5.5 is frustrating. I hate its character (there's a whiff of subtle disdain and friction), I don't think it is remotely as intellectually deep as Astra. It feels like it's built for normie work and demos on X.","url":"https://x.com/teortaxesTex/status/2104304901455012256","author":null,"stance":-2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"I am sorry, Opus 5.5 is nowhere near Fable 5.1 Planning to stick with Fable 5.1 for all hard-coding problems. It continues to be brilliant 🤷‍♀️","url":"https://x.com/bindureddy/status/2102500688047825354","author":null,"stance":-2,"hands_on":true,"basis":"used","verdict":"falls_short"},{"source":"x","text":"I gave Opus 5.5 a shot yesterday… it simply requires a lot of prompt iteration compared to Fable Net result - 2x my Fable bill not to mention my time Going back to Fable 5.1 🤷‍♀️","url":"https://x.com/bindureddy/status/2103761248030790052","author":null,"stance":-1,"hands_on":true,"basis":"used","verdict":"none"}]},"window":{"start":null,"end":null}}},{"model":"Gemini 3 Flash","score":84.5,"own_drop":true,"independent":false,"setting":"","source":{"drop":"Gemini 3 Flash","slug":"gemini-3-flash","url":"https://handyai.news/benchmarks/sentiment-bench#gemini-3-flash","published":"2025-12-17"},"launch_score":84.5,"reports":1,"maker":{"key":"google","name":"Google DeepMind","logo_url":"https://handyai.news/ui/makers/google.svg"},"open_weights":false,"released":"2025-12-17","dated_by":"own_drop","drop_url":"https://handyai.news/modeldrop/gemini-3-flash","rank":7,"sentiment":{"reception":86.1,"lives_up_to_benchmarks":79.6,"opinions":156,"hands_on":49,"crowd":85.8,"sources":{"x":{"score":78.2,"adjusted":74.8,"typical":76.9,"opinions":53,"positive":37,"negative":16},"hn":{"score":69.5,"adjusted":99.5,"typical":48.4,"opinions":86,"positive":61,"negative":26},"news":{"score":96.6,"adjusted":86.4,"typical":89.1,"opinions":14,"positive":14,"negative":0}},"reviewers":{"score":77.2,"adjusted":86.9,"count":2,"each":[{"name":"Peter Gostev","score":91.9,"url":"https://x.com/petergostev/status/2001339049320329621"},{"name":"Simon Willison","score":75,"url":"https://simonwillison.net/2025/Dec/17/gemini-3-flash/"}]},"verdicts":{"beats":6,"matches":5,"falls_short":7},"buzz":{"x_posts_that_week":null,"hn_points":1102,"articles":38},"quotes":{"positive":[{"source":"x","text":"It's absolutely insane how well Gemini 3 Flash works with our Chrome extension, you can design and prototype your REAL product in minutes now. Feels like a paradigm shift. pic.twitter.com/Or9bclcibU","url":"https://x.com/skirano/status/2001367818118582431","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"Gemini 3.0 Flash (testing) I was thankfully able to test Gemini 3.0 Flash in advance. It's a fantastic model! As expected, it's significantly faster, but the quality is approaching that of the Pro (the Pro is still better, of course, but the Flash isn't far behind). Here are a… p","url":"https://x.com/kimmonismus/status/2001324859100676514","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"I've had early access to gemini-3.0-flash for about a week. It's the closest we've gotten to \"intelligence too cheap/fast to meter\" yet, by a substantial margin. Some things I tested 🧵 https://t.co/jDzWTYKsSz","url":"https://x.com/adonis_singh/status/2001323749736604052","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"}],"negative":[{"source":"x","text":"Gemini 3 Flash with a crazy 91% Hallucination Rate on AA-Omniscience benchmark LisanBench hinted at this by showing lower average validity https://t.co/mtce5SUIe5 pic.twitter.com/HvDn0tPkAV","url":"https://x.com/scaling01/status/2001343898074145176","author":null,"stance":-2,"hands_on":false,"basis":"evidence","verdict":"none"},{"source":"x","text":"No the hell it doesn't. People need to stop this nonsense. Gemini 3 flash hallucination 100x worse and is not a top teor model. These benchmarks are broken and new tears are needed. Just Google gemini 3 flash hallucinations. It makes up bullshit 97% https://t.co/bixNXHEUhH","url":"https://x.com/ArcherNightfall/status/2002691584857722903","author":null,"stance":-2,"hands_on":false,"basis":"evidence","verdict":"none"},{"source":"x","text":"this validates the 91% hallucination rate of Gemini 3 Flash https://t.co/aAp2HeVTw7","url":"https://x.com/saurabhchopraa/status/2002728779328716955","author":null,"stance":-2,"hands_on":false,"basis":"evidence","verdict":"none"}]},"window":{"start":null,"end":null}}},{"model":"Composer","score":84.1,"own_drop":true,"independent":false,"setting":"","source":{"drop":"Composer","slug":"composer","url":"https://handyai.news/benchmarks/sentiment-bench#composer","published":"2025-10-29"},"launch_score":84.1,"reports":1,"maker":{"key":"cursor","name":"Cursor","logo_url":"https://handyai.news/ui/makers/cursor.svg"},"open_weights":false,"released":"2025-10-29","dated_by":"own_drop","drop_url":"https://handyai.news/modeldrop/composer","rank":8,"sentiment":{"reception":83.8,"lives_up_to_benchmarks":84.8,"opinions":72,"hands_on":42,"crowd":86.6,"sources":{"x":{"score":90.8,"adjusted":90.4,"typical":76.9,"opinions":46,"positive":40,"negative":6},"hn":{"score":61.5,"adjusted":89.5,"typical":48.4,"opinions":25,"positive":15,"negative":12},"news":{"score":null,"adjusted":null,"typical":89.1,"opinions":0,"positive":0,"negative":0}},"reviewers":{"score":71.3,"adjusted":77.3,"count":1,"each":[{"name":"Every","score":75,"url":"https://every.to/vibe-check/vibe-check-cursor-2-0-and-composer-1-alpha"}]},"verdicts":{"beats":1,"matches":1,"falls_short":4},"buzz":{"x_posts_that_week":null,"hn_points":215,"articles":6},"quotes":{"positive":[{"source":"x","text":"Composer is one of the craziest models I've ever used. Fast as hell and smart too. Edits multiple files at a time. Feels fundamentally different from other tools https://t.co/6q8j5mcMDV","url":"https://x.com/theo/status/1983602477556728156","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"Cursor team cooked incredibly hard with Composer 1... I'm more bearish than ever on Anthropic. Composer 1 is cheaper than Claude, can cook up beautiful frontends (like Claude, unlike Codex), and is so blazing fast. It's the first time that a coding model has me hooked in the…","url":"https://x.com/vasuman/status/1983722421099069483","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"cursor 2.0 - composer 1 is INSANELY fast (and accurate too now!) tip: add \"think through\" and it'll take time to ensure it takes the right approach then executes it in lightening speed pic.twitter.com/f5T24rsf4W","url":"https://x.com/amirmxt/status/1983630612616585616","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"}],"negative":[{"source":"x","text":"cursor's new composer-1 model it's surprisingly expensive... sonnet 4.5: - 680,375 total tokens ($0.57) - 4 minutes 10 seconds composer-1 - 784,430 total tokens ($0.36) - 1 minutes 37 seconds sounds good, right? wrong. sonnet does a much better job by doing additional work… https","url":"https://x.com/robinebers/status/1985255648616185857","author":null,"stance":-1,"hands_on":true,"basis":"used","verdict":"falls_short"},{"source":"hn","text":"I used the new system tonight and it felt like a definite downgrade. Generated a few non-working basic apps, couldn't handle CSS in a NextJS environment. Terminal context didn't work. And it went back to not reasoning through the problem until resolution. And kept slowing down. I","url":"https://news.ycombinator.com/item?id=45757214","author":"toobulkeh","stance":-2,"hands_on":true,"basis":"used","verdict":"falls_short"},{"source":"x","text":"cursor's composer got 32% on the nextjs evals i tried qwen3 with opencode on a fast inference provider it scored 36% and was 13% faster off the shelf, zero tuning and it's better...wtf?","url":"https://x.com/thdxr/status/1983753005439611358","author":null,"stance":-1,"hands_on":false,"basis":"evidence","verdict":"falls_short"}]},"window":{"start":null,"end":null}}},{"model":"DeepSeek V4.1 Flash 1017","score":83.9,"own_drop":true,"independent":false,"setting":"","source":{"drop":"DeepSeek V4.1 Flash 1017","slug":"deepseek-v41-flash-1017","url":"https://handyai.news/benchmarks/sentiment-bench#deepseek-v41-flash-1017","published":"2026-10-17"},"launch_score":83.9,"reports":1,"maker":{"key":"deepseek","name":"DeepSeek","logo_url":"https://handyai.news/ui/makers/deepseek.svg"},"open_weights":true,"released":"2026-10-17","dated_by":"own_drop","drop_url":"https://handyai.news/modeldrop/deepseek-v41-flash-1017","rank":9,"sentiment":{"reception":84.2,"lives_up_to_benchmarks":83,"opinions":143,"hands_on":80,"crowd":89.9,"sources":{"x":{"score":86.4,"adjusted":85,"typical":76.9,"opinions":46,"positive":37,"negative":10},"hn":{"score":82.6,"adjusted":100,"typical":48.4,"opinions":88,"positive":72,"negative":18},"news":{"score":94.9,"adjusted":83.4,"typical":89.1,"opinions":6,"positive":6,"negative":0}},"reviewers":{"score":67.4,"adjusted":70.9,"count":3,"each":[{"name":"Nathan Lambert","score":100,"url":"https://www.interconnects.ai/p/latest-open-artifacts-25-deepseek"},{"name":"Simon Willison","score":75,"url":"https://news.ycombinator.com/item?id=50130964"},{"name":"xlr8harder","score":25,"url":"https://x.com/xlr8harder/status/2111386674554892616"}]},"verdicts":{"beats":1,"matches":6,"falls_short":5},"buzz":{"x_posts_that_week":null,"hn_points":1669,"articles":17},"quotes":{"positive":[{"source":"x","text":"We are hitting levels of speed you can't imagine. Deepseek-v4.1-flash-1017 @ 675.9 tok/s c1 Recipes coming pic.twitter.com/WtBKn5uDkt","url":"https://x.com/0xSero/status/2113067168954982911","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"609 tok/s on deepseek-v4.1-flash-1017 has just been reached We are approaching Cerebras speeds at home https://t.co/HKIH5K40lk","url":"https://x.com/0xSero/status/2112877347539935540","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"You have Deepseek-v4.1-flash-1017 for free at amazing speeds and you're dooming? Opencode is going god's work. pic.twitter.com/qM5n5qBRhM","url":"https://x.com/0xSero/status/2113003769344704710","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"}],"negative":[{"source":"x","text":"This is how \"the new faster DeepSeek 4.1 Flash\" sees illustrations in documents. Green is the ground truth. Red is the cross-eyed DeepSeek. pic.twitter.com/vtJCWOfkV4","url":"https://x.com/burkov/status/2111708858280198162","author":null,"stance":-2,"hands_on":true,"basis":"used","verdict":"falls_short"},{"source":"x","text":"I just tested DeepSeek-V4.1-Flash-1017 on LisanBench I've previously suspected that labs are RL'ing on it, but for DeepSeek I am 100% certain that they do for example: it includes a chain of 64 two letter words I have only seen this behavior once when I hillclimbed LisanBench… ht","url":"https://x.com/scaling01/status/2111528125280464945","author":null,"stance":-1,"hands_on":true,"basis":"used","verdict":"falls_short"},{"source":"hn","text":"I feel like I'm an outlier based on all these comments. This is still working through my personal benchmarks, but it's spending 30% more time and performing worse than the previous 4.1 flash did. Benchmark is historical PR reviews. Models are put in a sandbox with a git and db sn","url":"https://news.ycombinator.com/item?id=50144401","author":"springtimesun","stance":-2,"hands_on":true,"basis":"used","verdict":"falls_short"}]},"window":{"start":null,"end":null}}},{"model":"Kimi K2.5","score":83.7,"own_drop":true,"independent":false,"setting":"","source":{"drop":"Kimi K2.5","slug":"kimi-k25","url":"https://handyai.news/benchmarks/sentiment-bench#kimi-k25","published":"2026-01-27"},"launch_score":83.7,"reports":1,"maker":{"key":"moonshot","name":"Moonshot AI","logo_url":"https://handyai.news/ui/makers/moonshot.svg"},"open_weights":true,"released":"2026-01-27","dated_by":"own_drop","drop_url":"https://handyai.news/modeldrop/kimi-k25","rank":10,"sentiment":{"reception":85.5,"lives_up_to_benchmarks":78.3,"opinions":135,"hands_on":61,"crowd":89,"sources":{"x":{"score":84,"adjusted":82,"typical":76.9,"opinions":49,"positive":39,"negative":10},"hn":{"score":83.1,"adjusted":100,"typical":48.4,"opinions":73,"positive":60,"negative":13},"news":{"score":96.1,"adjusted":85.5,"typical":89.1,"opinions":12,"positive":12,"negative":0}},"reviewers":{"score":71.3,"adjusted":77.3,"count":1,"each":[{"name":"Simon Willison","score":75,"url":"https://news.ycombinator.com/item?id=46777973"}]},"verdicts":{"beats":0,"matches":4,"falls_short":6},"buzz":{"x_posts_that_week":null,"hn_points":890,"articles":18},"quotes":{"positive":[{"source":"x","text":"Output quality: Kimi 2.5 > Opus 4.5 (Claude Code) > GPT 5.2 > Gemini 3 Pro CC's multiagent/subagent system is damn good. Format was great too, but it ranks below Kimi because it included irrelevant stuff in the output https://t.co/aJl1BtijXp","url":"https://x.com/zephyr_z9/status/2015952979158040667","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"Kimi K2.5 (Kimi CLI) vs MiniMax 2.1 (CC) vs GLM 4.7 (CC). 🔥 Same prompt to create a single-page website for \"PHANTOM PROTOCOL\" a fictional tactical shooter video game, 0-shot. Spoiler IMO: 🥇 Kimi K2.5 is another league 🥈 MiniMax 2.1 🥉 GLM 4.7 pic.twitter.com/C4R7j2IobW","url":"https://x.com/ivanfioravanti/status/2017486165188690310","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"Kimi-K2.5 can use a computer. It ACES browser user, look at that smooth navigation. Ok computer. pic.twitter.com/eVHbYmgRpg","url":"https://x.com/0xSero/status/2016266795355648489","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"}],"negative":[{"source":"x","text":"kimi k2.5 is overhyped. compare the reference design i gave it to the generated result pic.twitter.com/ruNf0HC7N0","url":"https://x.com/kikiolaola/status/2016655098374713568","author":null,"stance":-2,"hands_on":true,"basis":"used","verdict":"falls_short"},{"source":"x","text":"K2.5 - GLM-4.7 and MiniMax M2.1 solving a Rubik's Cube in 3D. 👀 Kimi CLI and Claude Code used here. 🥇 GLM-4.7 36 secs and perfect! 🥈 K2.5 - Missing colors in the cube 🥉 MiniMax 2.1 - Missing colors and failed to autosolve pic.twitter.com/PEuu2yqAsE","url":"https://x.com/ivanfioravanti/status/2017595857986527570","author":null,"stance":-1,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"initial impressions of Kimi K2.5: it needs a Speciale/Heavy version or agent straitjacket to shine it's too lazy a thinker. 5.2 and Speciale destroy it on Ph.D level tasks where they go for >>20K thinking tokens Per token, it's frontier level Surprisingly good vision too","url":"https://x.com/teortaxesTex/status/2016009804259487850","author":null,"stance":-1,"hands_on":true,"basis":"used","verdict":"none"}]},"window":{"start":null,"end":null}}},{"model":"Qwen3.8","score":83.6,"own_drop":true,"independent":false,"setting":"","source":{"drop":"Qwen3.8","slug":"qwen38","url":"https://handyai.news/benchmarks/sentiment-bench#qwen38","published":"2026-08-12"},"launch_score":83.6,"reports":1,"maker":{"key":"qwen","name":"Alibaba","logo_url":"https://handyai.news/ui/makers/qwen.svg"},"open_weights":true,"released":"2026-08-12","dated_by":"own_drop","drop_url":"https://handyai.news/modeldrop/qwen38","rank":11,"sentiment":{"reception":87.7,"lives_up_to_benchmarks":71.4,"opinions":218,"hands_on":108,"crowd":87.7,"sources":{"x":{"score":81.3,"adjusted":78.6,"typical":76.9,"opinions":46,"positive":35,"negative":12},"hn":{"score":73.2,"adjusted":100,"typical":48.4,"opinions":156,"positive":103,"negative":53},"news":{"score":96.6,"adjusted":86.4,"typical":89.1,"opinions":11,"positive":11,"negative":0}},"reviewers":{"score":77.8,"adjusted":87.8,"count":1,"each":[{"name":"Simon Willison","score":91.4,"url":"https://x.com/simonw/status/2089112517796827439"}]},"verdicts":{"beats":0,"matches":5,"falls_short":9},"buzz":{"x_posts_that_week":null,"hn_points":2532,"articles":21},"quotes":{"positive":[{"source":"x","text":"This is what it looks like to run Qwen 3.8 27B on a 5090 and get more than 100 tok/s. 6-month-old frontier AI, now running locally on a gaming GPU. pic.twitter.com/DdGq29W2Mv","url":"https://x.com/Hesamation/status/2089497073389347025","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"This is unreal. Qwen 3.8 27b running locally on my RTX 5090 BEATS Opus 4.8 in my benchmarks (funny enough slightly loses to Opus 4.6 lol) Speeds are just as good if not sometimes MUCH better. Got up to 200 tk/s in some tests We live in a new world. Unlimited, free super… pic.twit","url":"https://x.com/AlexFinn/status/2089455600338116980","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"I tested Gemini 3.7 Flash vs Qwen 3.8 vs Grok 4.6 vs GLM 5.3 Same prompt, 2 scenes: a harvester and farmers working. honestly? Qwen 3.8 impressed me. it just did what I asked. Grok 4.6 failed on coloring. Qwen>GLM> Grok> Gemini which one did better for you? pic.twitter.com/kCN7zG","url":"https://x.com/Bhavani_00007/status/2088646139310645675","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"}],"negative":[{"source":"x","text":"literally just tried to let qwen3.8 27b one shot a fluid simulation webapp it thought for 40k tokens, took around 1 hour to generate and it's just black and doesn't work meanwhile Opus 4.5 just oneshots it in a minute and works wonderfully exact same prompt https://t.co/wp9nYD4DH","url":"https://x.com/scaling01/status/2089784644400976254","author":null,"stance":-2,"hands_on":true,"basis":"used","verdict":"falls_short"},{"source":"x","text":"i measured qwen 3.8 27b dense, raw on a single RTX 3090 24gb GPU, before the hype finished loading. here is the honest read: 25.4 tok/s on the exact card and test where 3.6 ran 40tok/s. slower on day one, and i'm posting it anyway, because the rest of the data says watch this… ht","url":"https://x.com/sudoingX/status/2088327367500714279","author":null,"stance":-1,"hands_on":true,"basis":"used","verdict":"none"},{"source":"hn","text":"So far what I am seeing in my seemingly simple \"Wordpress last-login plugin\" test is that in xhigh reasoning mode (the default, seemingly) it overthinks so badly that it writes terrible bushy code with edge cases caused by going down rabbit holes. In \"medium\" reasoning mode, you ","url":"https://news.ycombinator.com/item?id=49309783","author":"dofm","stance":-2,"hands_on":true,"basis":"used","verdict":"none"}]},"window":{"start":null,"end":null}}},{"model":"Gemini 3 Pro","score":82.9,"own_drop":true,"independent":false,"setting":"","source":{"drop":"Gemini 3 Pro","slug":"gemini-3-pro","url":"https://handyai.news/benchmarks/sentiment-bench#gemini-3-pro","published":"2025-11-18"},"launch_score":82.9,"reports":1,"maker":{"key":"google","name":"Google DeepMind","logo_url":"https://handyai.news/ui/makers/google.svg"},"open_weights":false,"released":"2025-11-18","dated_by":"own_drop","drop_url":"https://handyai.news/modeldrop/gemini-3-pro","rank":12,"sentiment":{"reception":87.9,"lives_up_to_benchmarks":68,"opinions":179,"hands_on":97,"crowd":85.5,"sources":{"x":{"score":83.8,"adjusted":81.7,"typical":76.9,"opinions":41,"positive":33,"negative":8},"hn":{"score":63.7,"adjusted":92.3,"typical":48.4,"opinions":106,"positive":58,"negative":49},"news":{"score":94.1,"adjusted":82,"typical":89.1,"opinions":27,"positive":25,"negative":2}},"reviewers":{"score":81.3,"adjusted":93.5,"count":5,"each":[{"name":"Andrej Karpathy","score":100,"url":"https://x.com/karpathy/status/1990854771058913347"},{"name":"Ethan Mollick","score":100,"url":"https://www.oneusefulthing.org/p/three-years-from-gpt-3-to-gemini"},{"name":"Simon Willison","score":75,"url":"https://simonwillison.net/2025/Nov/18/gemini-3/"},{"name":"Every","score":75,"url":"https://every.to/vibe-check/vibe-check-gemini-3-pro-a-reliable-workhorse-with-surprising-flair"},{"name":"Zvi Mowshowitz","score":75,"url":"https://thezvi.substack.com/p/gemini-3-model-card-and-safety-framework"}]},"verdicts":{"beats":11,"matches":11,"falls_short":18},"buzz":{"x_posts_that_week":null,"hn_points":2529,"articles":60},"quotes":{"positive":[{"source":"x","text":"Gemini 3 is truly a great model. It created the Rubik's cube simulation on the first try, in seconds, with ~300 lines of code. Flawless. pic.twitter.com/kqqttn3e7t","url":"https://x.com/MatthewBerman/status/1990832167635857415","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"matches"},{"source":"x","text":"My notes on Gemini 3 # Short Version: First of all: you've all seen the benchmarks, so I don't think you need me to judge this one. Still, based on my tests, this is as real as it gets, and I want to talk about it. This model outperforms GPT-5 Pro, Gemini 2.5 Deep Think, and… pic","url":"https://x.com/VictorTaelin/status/1990844923994886282","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"matches"},{"source":"x","text":"I used Gemini 3 to build a rough iOS app for competitive sleep tracking - built on the first try with zero bugs. Very impressive - other models have half a dozen issues that need debugging before they can be run on the simulator. Gemini got it working on the first try. pic.twitte","url":"https://x.com/AiBreakfast/status/1991190478210101307","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"matches"}],"negative":[{"source":"x","text":"I am happy to report Gemini 3 continues to have the suicidal tendencies of its predecessor LMAO pic.twitter.com/sE9Cffw5Hw","url":"https://x.com/synthwavedd/status/1991236328621576651","author":null,"stance":-2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"AAAA I'VE LOST IT We are in a fucking AI winter for real, frog is right Google has not been able to fix this retarded “hypothetical future date” gaslighting. For Gemini 3 Pro, today is still early 2024. https://t.co/LoJABAywzx pic.twitter.com/ESjicms3gT","url":"https://x.com/teortaxesTex/status/1990900751116259428","author":null,"stance":-2,"hands_on":false,"basis":"evidence","verdict":"falls_short"},{"source":"x","text":"your goal now is to review the PARSER. remember: - do NOT complete the file - do NOT write ANY code just answer this question, in ENGLISH: Q: is the parser correct and following the spec? (30 mins later) Gemini 3: no, because it is missing the rest of the file. here it is:","url":"https://x.com/VictorTaelin/status/1992612415792644096","author":null,"stance":-1,"hands_on":true,"basis":"used","verdict":"none"}]},"window":{"start":null,"end":null}}},{"model":"Grok 4.5","score":82.5,"own_drop":true,"independent":false,"setting":"","source":{"drop":"Grok 4.5","slug":"grok-45","url":"https://handyai.news/benchmarks/sentiment-bench#grok-45","published":"2026-07-08"},"launch_score":82.5,"reports":1,"maker":{"key":"xai","name":"SpaceXAI","logo_url":"https://handyai.news/ui/makers/xai.svg"},"open_weights":false,"released":"2026-07-09","dated_by":"own_drop","drop_url":"https://handyai.news/modeldrop/grok-45","rank":13,"sentiment":{"reception":81.8,"lives_up_to_benchmarks":84.7,"opinions":145,"hands_on":33,"crowd":90.3,"sources":{"x":{"score":99.2,"adjusted":100,"typical":76.9,"opinions":34,"positive":34,"negative":0},"hn":{"score":58.5,"adjusted":85.8,"typical":48.4,"opinions":96,"positive":53,"negative":52},"news":{"score":91,"adjusted":76.5,"typical":89.1,"opinions":12,"positive":11,"negative":1}},"reviewers":{"score":61.8,"adjusted":61.8,"count":3,"each":[{"name":"Sebastian Raschka","score":75,"url":"https://x.com/rasbt/status/2075982283509571666"},{"name":"Max Woolf","score":75,"url":"https://news.ycombinator.com/item?id=48883195"},{"name":"Ethan Mollick","score":25,"url":"https://x.com/emollick/status/2075076753903558997"}]},"verdicts":{"beats":4,"matches":3,"falls_short":7},"buzz":{"x_posts_that_week":null,"hn_points":1108,"articles":35},"quotes":{"positive":[{"source":"x","text":"grok 4.5 is the first model that i've found a number of easy ways to make quick money, and no small amount. large language models are weird. but in a way, this is a good benchmark.","url":"https://x.com/iruletheworldmo/status/2076314051701789097","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"This is literally my new workflow now: Realtime Research → Grok 4.5 High Planning & Orchestration→ Fable 5 Max / XHigh Day-to-day Coding/Debug → Grok 4.5 High Write & Run Tests → Grok 4.5 High Complex Coding/Debug → GPT-5.6 Sol XHigh Frontend → Fable 5 High Bookmark this","url":"https://x.com/minchoi/status/2076326073374023926","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"Grok 4.5 has a new surprise every day! It literally is the BEST MODEL for pdf creation Yes, it beats everyone - sol, opus, fable etc.","url":"https://x.com/bindureddy/status/2076899594810896799","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"beats"}],"negative":[{"source":"hn","text":"So strange to write a whole post with Claude giving the best results and Grok consistently the worst, but awarding Grok the winner because at least it did the worst fastest?","url":"https://news.ycombinator.com/item?id=48839687","author":"jeffgreco","stance":-2,"hands_on":false,"basis":"evidence","verdict":"none"},{"source":"hn","text":"Grok failed the Rubik's cube. I pressed Scramble twice and then solve and it didn't solve the cube. Opus did.","url":"https://news.ycombinator.com/item?id=48841214","author":"JoyfulPanda","stance":-2,"hands_on":true,"basis":"used","verdict":"falls_short"},{"source":"hn","text":"Grok doesn't even solve the cube if you scramble more than once, that is NOT the AI you wan't building your software, no matter how cheap.","url":"https://news.ycombinator.com/item?id=48844100","author":"sceptic123","stance":-2,"hands_on":false,"basis":"evidence","verdict":"falls_short"}]},"window":{"start":null,"end":null}}},{"model":"Kimi K2.7 Code","score":82.5,"own_drop":true,"independent":false,"setting":"","source":{"drop":"Kimi K2.7 Code","slug":"kimi-k27-code","url":"https://handyai.news/benchmarks/sentiment-bench#kimi-k27-code","published":"2026-06-12"},"launch_score":82.5,"reports":1,"maker":{"key":"moonshot","name":"Moonshot AI","logo_url":"https://handyai.news/ui/makers/moonshot.svg"},"open_weights":true,"released":"2026-06-12","dated_by":"own_drop","drop_url":"https://handyai.news/modeldrop/kimi-k27-code","rank":14,"sentiment":{"reception":84.1,"lives_up_to_benchmarks":77.8,"opinions":73,"hands_on":35,"crowd":84.1,"sources":{"x":{"score":79.6,"adjusted":76.5,"typical":76.9,"opinions":49,"positive":38,"negative":11},"hn":{"score":71.2,"adjusted":100,"typical":48.4,"opinions":20,"positive":11,"negative":9},"news":{"score":null,"adjusted":null,"typical":89.1,"opinions":4,"positive":4,"negative":0}},"reviewers":{"score":null,"adjusted":null,"count":0,"each":[]},"verdicts":{"beats":0,"matches":1,"falls_short":4},"buzz":{"x_posts_that_week":null,"hn_points":463,"articles":11},"quotes":{"positive":[{"source":"x","text":"我要强烈推荐 Kimi K2.7 High Speed (高速版）！太神奇了！ Fable 5只活了3天就下架了，那个「墨流」Demo我见一次就再也忘不了。 我让Opus 4.8(左)和 Kimi K2.7 High Speed(右)同时复刻它： 两边都成功，质量上K2.7只略逊于Opus 4.8，可这个速度！ Opus花了32分钟，K2.7 只用了11分钟！ 看视频！ pic.twitter.com/ZrjT00TYmd","url":"https://x.com/bourneliu66/status/2066436537085276284","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"🚨 Kimi Cooked So Hard with this release This is one Shotted by Kimi k2.7 fast , math guys try kimi i think its better in math too now Fable was able to do this but kimi only took less than a min lol , i had early access to it https://t.co/ffZd9DBW3y pic.twitter.com/oUmWqu9Q18","url":"https://x.com/chetaslua/status/2066504396666794117","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"based on everything that has happened over the last week (and year really) and how good k2.7 is in this harness, it is getting tempting to make this available to y'all pic.twitter.com/4oOn5ADe21","url":"https://x.com/xjdr/status/2065855561552588895","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"}],"negative":[{"source":"x","text":"GLM 5.2 is the best Chinese open source model. Kimi K2.7 Code does not even come close. GLM 5.2 ranks #1 on BridgeBench reasoning while Kimi K2.7 code ranks #11. Kimi K2.7 even saw a REGRESSION from Kimi K2.6 on the reasoning benchmark. pic.twitter.com/QgQRk17ZM3","url":"https://x.com/bridgemindai/status/2066608105434010063","author":null,"stance":-2,"hands_on":false,"basis":"evidence","verdict":"none"},{"source":"x","text":"Kimi K2.7 Code is a really bad coding model. My rules explicitly say not to launch Playwright or Chrome, yet it still tries to open ports and start browser-related processes. No idea why. I don't get the hype — it's terrible at following instructions and keeps ignoring my rules.","url":"https://x.com/OmedVibeCodes/status/2066993807380418681","author":null,"stance":-2,"hands_on":true,"basis":"used","verdict":"falls_short"},{"source":"x","text":"The Kimi K2.7 code is underwhelming. I attempted to port my JavaScript application to Kotlin, but it is far from finished; the model is overconfident and fails to recognize its own limitations.","url":"https://x.com/syqrel/status/2066604203561406568","author":null,"stance":-2,"hands_on":true,"basis":"used","verdict":"falls_short"}]},"window":{"start":null,"end":null}}},{"model":"GPT-5.5","score":82.5,"own_drop":true,"independent":false,"setting":"","source":{"drop":"GPT-5.5","slug":"gpt-55","url":"https://handyai.news/benchmarks/sentiment-bench#gpt-55","published":"2026-04-23"},"launch_score":82.5,"reports":1,"maker":{"key":"openai","name":"OpenAI","logo_url":"https://handyai.news/ui/makers/openai.svg"},"open_weights":false,"released":"2026-04-23","dated_by":"own_drop","drop_url":"https://handyai.news/modeldrop/gpt-55","rank":15,"sentiment":{"reception":84,"lives_up_to_benchmarks":78.2,"opinions":194,"hands_on":67,"crowd":80.9,"sources":{"x":{"score":85.1,"adjusted":83.4,"typical":76.9,"opinions":44,"positive":36,"negative":8},"hn":{"score":50.7,"adjusted":76.1,"typical":48.4,"opinions":116,"positive":58,"negative":59},"news":{"score":95.1,"adjusted":83.8,"typical":89.1,"opinions":25,"positive":24,"negative":1}},"reviewers":{"score":79.8,"adjusted":91.1,"count":7,"each":[{"name":"Peter Gostev","score":100,"url":"https://x.com/petergostev/status/2047379246012637577"},{"name":"Ethan Mollick","score":100,"url":"https://www.oneusefulthing.org/p/sign-of-the-future-gpt-55"},{"name":"Every","score":100,"url":"https://every.to/vibe-check/gpt-5-5"},{"name":"Simon Willison","score":100,"url":"https://simonwillison.net/2026/Apr/23/gpt-5-5/"},{"name":"Zvi Mowshowitz","score":75,"url":"https://thezvi.substack.com/p/gpt-55-the-system-card"},{"name":"xlr8harder","score":75,"url":"https://x.com/xlr8harder/status/2048984389426503831"},{"name":"Max Woolf","score":25,"url":"https://news.ycombinator.com/item?id=47879193"}]},"verdicts":{"beats":8,"matches":6,"falls_short":9},"buzz":{"x_posts_that_week":null,"hn_points":1996,"articles":60},"quotes":{"positive":[{"source":"x","text":"Just told GPT-5.5 to port and compile the codex app to Linux. 1. WTF. It actually worked. 2. WTF. That thing is incredible.","url":"https://x.com/Yampeleg/status/2049082782677164261","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"beats"},{"source":"x","text":"HOLLLLY SHIIIIIT 😳 GPT 5.5 xhigh in codex made this paper physics website with wind effect one shot check the physics , ui and interaction @sama you guys cooked and this is my first time i can suggest chatgpt is go to ai solution for everything https://t.co/eA2Rkyt1wO pic.twitte","url":"https://x.com/chetaslua/status/2047393912122486845","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"beats"},{"source":"x","text":"GPT 5.5 is much smarter than I thought Yesterday, I did one-shots, coding, benchmarks, and was disappointed. Today, I did it all again, except via the API, which is now available. Results changed completely: → one-shot prompts went from bad to very good → excellent coding… pic.tw","url":"https://x.com/VictorTaelin/status/2047818978664268071","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"beats"}],"negative":[{"source":"x","text":"GPT-5.5 hallucinates on 86% of non correct responses. GPT-5.4 was 89%. A 3 point improvement. On a metric where Claude Opus 4.7 sits at 36%. Two years of research. Trained on Stargate. Still hallucinates more than every Anthropic model. The moat isn't gone. It just got… pic.twitt","url":"https://x.com/bridgemindai/status/2047413750844788741","author":null,"stance":-2,"hands_on":false,"basis":"evidence","verdict":"none"},{"source":"x","text":"GPT 5.5 is terrible at frontend design. Ranked #11 on Design Arena. Way behind Claude Opus 4.7 and even Kimi K2.6. OpenAI told us that GPT 5.5 would be way better at UI. They lied. pic.twitter.com/wyfyJYfqoQ","url":"https://x.com/bridgemindai/status/2048744421395407294","author":null,"stance":-2,"hands_on":false,"basis":"evidence","verdict":"falls_short"},{"source":"x","text":"$5 per mil in, $30 per mil out. GPT-5.5 is smart. I've been using it for a bit. It's also weird, hard to wrangle, and too expensive IMO. Double the price of GPT-5.4. 20% more expensive than Opus 4.7. https://t.co/C8k68quwtw","url":"https://x.com/theo/status/2047379285107691835","author":null,"stance":-1,"hands_on":true,"basis":"used","verdict":"none"}]},"window":{"start":null,"end":null}}},{"model":"Composer 2.5","score":82.3,"own_drop":true,"independent":false,"setting":"","source":{"drop":"Composer 2.5","slug":"composer-25","url":"https://handyai.news/benchmarks/sentiment-bench#composer-25","published":"2026-05-18"},"launch_score":82.3,"reports":1,"maker":{"key":"cursor","name":"Cursor","logo_url":"https://handyai.news/ui/makers/cursor.svg"},"open_weights":false,"released":"2026-05-18","dated_by":"own_drop","drop_url":"https://handyai.news/modeldrop/composer-25","rank":16,"sentiment":{"reception":81.8,"lives_up_to_benchmarks":83.7,"opinions":97,"hands_on":39,"crowd":83.7,"sources":{"x":{"score":93,"adjusted":93.1,"typical":76.9,"opinions":46,"positive":40,"negative":6},"hn":{"score":46.8,"adjusted":71.2,"typical":48.4,"opinions":40,"positive":19,"negative":24},"news":{"score":95.6,"adjusted":84.6,"typical":89.1,"opinions":10,"positive":10,"negative":0}},"reviewers":{"score":71.3,"adjusted":77.3,"count":1,"each":[{"name":"Every","score":75,"url":"https://x.com/kieranklaassen/status/2057951067682521345"}]},"verdicts":{"beats":4,"matches":2,"falls_short":2},"buzz":{"x_posts_that_week":null,"hn_points":308,"articles":14},"quotes":{"positive":[{"source":"x","text":"Composer 2.5 fixed my ADHD coding. It's so fast, I don't have time to do something else. And it's good enough for most tasks, so it's my new default now.","url":"https://x.com/marclou/status/2057307977863041454","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"just used composer 2.5 for the first time and it's ridiculously good. can only assume people are incentivized to hype other models because this is incredible.","url":"https://x.com/iruletheworldmo/status/2057170480566538277","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"Cursor ( @cursor_ai ) really shouldn't be underestimated. I'm testing Composer 2.5 right now after coming back to Cursor for the first time in a while, and it has clearly improved. The whole experience feels smoother, more polished, and more coherent than I remembered. And the… p","url":"https://x.com/mark_k/status/2057140486314549485","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"}],"negative":[{"source":"x","text":"Tried Composer 2.5 on a semi-complicated task, the result is far worse than Opus 4.7 High or GPT 5.5 High. But it was fast.","url":"https://x.com/tdinh_me/status/2057782057103958485","author":null,"stance":-1,"hands_on":true,"basis":"used","verdict":"falls_short"},{"source":"hn","text":"I tested it yesterday. It is pretty bad. Just like with Composer 2, it's fast, but quality is nowhere near what Cursor claims with their benchmarks. It is not even at Opus 4.5 level. I gave it a mix of refactoring tasks and new feature tasks. For each one, I had it write a plan, ","url":"https://news.ycombinator.com/item?id=48193131","author":"enraged_camel","stance":-2,"hands_on":true,"basis":"used","verdict":"falls_short"},{"source":"hn","text":"I want to like composer, but I just can't. - Its communication style is completely opposite to Anthropic models. It's not as bad as OpenAI's models, which are obsessed with \"shapes\", \"wrinkles\", hyphenated-words, and other cryptic formulations that make you feel like you're not o","url":"https://news.ycombinator.com/item?id=48192269","author":"jorl17","stance":-2,"hands_on":true,"basis":"used","verdict":"none"}]},"window":{"start":null,"end":null}}},{"model":"Kimi K2.6","score":82.3,"own_drop":true,"independent":false,"setting":"","source":{"drop":"Kimi K2.6","slug":"kimi-k26","url":"https://handyai.news/benchmarks/sentiment-bench#kimi-k26","published":"2026-04-20"},"launch_score":82.3,"reports":1,"maker":{"key":"moonshot","name":"Moonshot AI","logo_url":"https://handyai.news/ui/makers/moonshot.svg"},"open_weights":true,"released":"2026-04-22","dated_by":"own_drop","drop_url":"https://handyai.news/modeldrop/kimi-k26","rank":17,"sentiment":{"reception":85.4,"lives_up_to_benchmarks":73.2,"opinions":118,"hands_on":50,"crowd":88.1,"sources":{"x":{"score":84.3,"adjusted":82.4,"typical":76.9,"opinions":55,"positive":46,"negative":9},"hn":{"score":68.1,"adjusted":97.7,"typical":48.4,"opinions":51,"positive":39,"negative":12},"news":{"score":95.4,"adjusted":84.3,"typical":89.1,"opinions":10,"positive":10,"negative":0}},"reviewers":{"score":72.3,"adjusted":78.9,"count":2,"each":[{"name":"Peter Gostev","score":75,"url":"https://x.com/petergostev/status/2046619960672506127"},{"name":"Simon Willison","score":75,"url":"https://news.ycombinator.com/item?id=47837045"}]},"verdicts":{"beats":1,"matches":10,"falls_short":7},"buzz":{"x_posts_that_week":null,"hn_points":761,"articles":16},"quotes":{"positive":[{"source":"x","text":"We are entering an extremely exciting era for open-weight models. Kimi K2.6 now feels like a top agentic model. I took it for a spin via @FireworksAI_HQ fast inference APIs. Kimi K2.6 has impressive agentic capabilities, design skills, and the ability to synthesize large… pic.twi","url":"https://x.com/omarsar0/status/2046682706210422874","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"matches"},{"source":"x","text":"Holy Shit Kimi Cooked 🐐 I have been testing kimi k2.6 for the last 7 days I have lots of cool demos to show you guys , kimi is keeping open-source alive Aesthetic: Beautiful front-end design with rich interaction https://t.co/ewd7ySuzfY pic.twitter.com/625TTi3lQt","url":"https://x.com/chetaslua/status/2046261605051290108","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"matches"},{"source":"x","text":"🚨Kimi K2.6 creates websites with the taste Tbh no body is talking about this , that it can calls image/video generation APIs, creating standout hero sections and visuals here is one example , many more coming https://t.co/CZp2sucAqK pic.twitter.com/uO6CSaOj96","url":"https://x.com/chetaslua/status/2046993315649004022","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"}],"negative":[{"source":"x","text":"KIMI K2.6 IS BENCHMAXED. I ran the BridgeBench lava lamp test on four models. GLM 5.1: pixel cube in a column. GPT-5.4: static bubbles. No flow. Kimi K2.6: a yellow rectangle. Claude Opus 4.7: an actual lava lamp. Chinese models ace benchmarks. Claude ships software. pic.twitter.","url":"https://x.com/bridgemindai/status/2046313533743468993","author":null,"stance":-2,"hands_on":true,"basis":"used","verdict":"falls_short"},{"source":"x","text":"GLM-5.1 vs Kimi K2.6 - Testing frontend and web design (not agent mode) GLM-5.1 definitely wins over Kimi here in my opinion. Interestingly they both used the same hero section main headline phrase. 1st video: GLM-5.1 2nd video: Kimi K2.6 Thinking https://t.co/BUkHXmfY50 pic.twit","url":"https://x.com/Lentils80/status/2046340087810564561","author":null,"stance":-1,"hands_on":true,"basis":"used","verdict":"none"},{"source":"hn","text":"Running it through opencode to their API and... it definitely seems like it's \"overthinking\" -- watching the thought process, it's been going for pages and pages and pages diagnosing and \"thinking\" things through... without doing anything. Sitting at 50k+ output tokens used now j","url":"https://news.ycombinator.com/item?id=47837089","author":"cmrdporcupine","stance":-2,"hands_on":true,"basis":"used","verdict":"falls_short"}]},"window":{"start":null,"end":null}}},{"model":"Gemma 4","score":82.3,"own_drop":true,"independent":false,"setting":"","source":{"drop":"Gemma 4","slug":"gemma-4","url":"https://handyai.news/benchmarks/sentiment-bench#gemma-4","published":"2026-04-02"},"launch_score":82.3,"reports":1,"maker":{"key":"google","name":"Google DeepMind","logo_url":"https://handyai.news/ui/makers/google.svg"},"open_weights":true,"released":"2026-04-02","dated_by":"own_drop","drop_url":"https://handyai.news/modeldrop/gemma-4","rank":18,"sentiment":{"reception":85.3,"lives_up_to_benchmarks":73.1,"opinions":196,"hands_on":81,"crowd":89,"sources":{"x":{"score":93.8,"adjusted":94.1,"typical":76.9,"opinions":45,"positive":39,"negative":6},"hn":{"score":69.4,"adjusted":99.4,"typical":48.4,"opinions":131,"positive":89,"negative":43},"news":{"score":81.1,"adjusted":59.1,"typical":89.1,"opinions":13,"positive":11,"negative":2}},"reviewers":{"score":71.1,"adjusted":76.9,"count":6,"each":[{"name":"xlr8harder","score":100,"url":"https://x.com/xlr8harder/status/2040311650952106247"},{"name":"Simon Willison","score":80,"url":"https://news.ycombinator.com/item?id=47617370"},{"name":"Ethan Mollick","score":75,"url":"https://x.com/emollick/status/2040851723774808310"},{"name":"Nathan Lambert","score":75,"url":"https://www.interconnects.ai/p/gemma-4-and-what-makes-an-open-model"},{"name":"Max Woolf","score":75,"url":"https://news.ycombinator.com/item?id=47616578"},{"name":"Peter Gostev","score":25,"url":"https://x.com/petergostev/status/2041525479153262899"}]},"verdicts":{"beats":7,"matches":17,"falls_short":23},"buzz":{"x_posts_that_week":null,"hn_points":3087,"articles":60},"quotes":{"positive":[{"source":"x","text":"Gemma 4 just dropped. I had it captioning video in real-time within an hour. Running locally on a MacBook. No cloud. No API. Real-time scene understanding. Oh and SAM3 is segmenting every object in the same frame. Same laptop. pic.twitter.com/xSYCMDY22y","url":"https://x.com/MaziyarPanahi/status/2040059349750464721","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"matches"},{"source":"x","text":"Watch Gemma-4-E4B casually identify sea animals by classifying images in a single agentic session using its vision capabilities. (impressive for a 4B model 🚀) I'm convinced: the agents of tomorrow are local, free, fast, and run on every computer! pic.twitter.com/mHPMiPPNSZ","url":"https://x.com/victormustar/status/2040711407902040128","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"matches"},{"source":"x","text":"this is Gemma 4 running locally on a 3 year old mac meaning: - free (=$0 no matter how much you use) - safe (you're not leaking all your data via unsafe APIs) - fast (as you can see) https://t.co/yX3gwJ3LF6","url":"https://x.com/ClementDelangue/status/2039791984827593053","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"matches"}],"negative":[{"source":"hn","text":"Prompt: > what is the Unix timestamp for this: 2026-04-01T16:00:00Z Qwen 3.5-27b-dwq > Thought for 8 minutes 34 seconds. 7074 tokens. > The Unix timestamp for 2026-04-01T16:00:00Z is: > 1775059200 (my comment: Wednesday, 1 April 2026 at 16:00:00) Gemma-4-26b-a4b > Thought for 33.","url":"https://news.ycombinator.com/item?id=47619651","author":"neonstatic","stance":-1,"hands_on":true,"basis":"used","verdict":"falls_short"},{"source":"hn","text":"I gave the same prompt (a small rust project that's not easy, but not overly sophisticated) to both Gemma-4 26b and Qwen 3.5 27b via OpenCode. Qwen 3.5 ran for a bit over an hour before I killed it, Gemma 4 ran for about 20 minutes before it gave up. Lots of failed tool calls. I ","url":"https://news.ycombinator.com/item?id=47619184","author":"swalsh","stance":-1,"hands_on":true,"basis":"used","verdict":"falls_short"},{"source":"hn","text":"So many comments, but in the end - it can't write a simple set of unit tests in go with mockery.","url":"https://news.ycombinator.com/item?id=47626622","author":"konart","stance":-2,"hands_on":true,"basis":"used","verdict":"falls_short"}]},"window":{"start":null,"end":null}}},{"model":"GPT-6.1 Sol","score":80.9,"own_drop":true,"independent":false,"setting":"","source":{"drop":"GPT-6.1 Sol","slug":"gpt-61-sol","url":"https://handyai.news/benchmarks/sentiment-bench#gpt-61-sol","published":"2026-09-29"},"launch_score":80.9,"reports":1,"maker":{"key":"openai","name":"OpenAI","logo_url":"https://handyai.news/ui/makers/openai.svg"},"open_weights":false,"released":"2026-09-29","dated_by":"own_drop","drop_url":"https://handyai.news/modeldrop/gpt-61-sol","rank":19,"sentiment":{"reception":80.7,"lives_up_to_benchmarks":81.7,"opinions":121,"hands_on":38,"crowd":76.4,"sources":{"x":{"score":73.8,"adjusted":69.4,"typical":76.9,"opinions":45,"positive":28,"negative":17},"hn":{"score":55.3,"adjusted":81.8,"typical":48.4,"opinions":69,"positive":36,"negative":34},"news":{"score":94.5,"adjusted":82.7,"typical":89.1,"opinions":5,"positive":5,"negative":0}},"reviewers":{"score":79.5,"adjusted":90.6,"count":2,"each":[{"name":"Peter Gostev","score":100,"url":"https://x.com/petergostev/status/2106765604711739623"},{"name":"Max Woolf","score":75,"url":"https://news.ycombinator.com/item?id=49896762"}]},"verdicts":{"beats":2,"matches":6,"falls_short":4},"buzz":{"x_posts_that_week":null,"hn_points":1147,"articles":60},"quotes":{"positive":[{"source":"x","text":"Updated run just finished. GPT-6.1 Sol still crushes. Turns out it performs WAY better in Codex than in mini-swe (what Artificial Analysis uses) https://t.co/dr11vvGtW3 pic.twitter.com/nNj9JcQQvK","url":"https://x.com/theo/status/2105063015712465094","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"beats"},{"source":"x","text":"🚨 Codex users GPT-6.1 Sol just got MUCH faster I tested the speed of GPT-6.1 Sol after OpenAI fixed the slowdown, and the results are honestly insane GPT-6.1 Sol is now up to ~20% faster than it was around DevDay The model quality was already crazy now the speed is finally… pic.","url":"https://x.com/afaa_him/status/2106252914503176434","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"GPT-6.1 Sol is a huge improvement on ARC-AGI: https://t.co/Bek8weIpPi pic.twitter.com/c6YwPeFcKo","url":"https://x.com/mark_k/status/2105560143765107186","author":null,"stance":2,"hands_on":false,"basis":"evidence","verdict":"none"}],"negative":[{"source":"x","text":"I hate that I need to prompt 6.1-Sol four more times to get the same information Opus gave me the first time","url":"https://x.com/adonis_singh/status/2106822486277173638","author":null,"stance":-1,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"Holy sh*t, Anthropic's subscriptions offer so much more value, even compared to the new and efficient GPT-6.1 Sol. It's not even close. SemiAnalysis puts Opus 5.5 at over 5x the API-equivalent value of GPT-6.1 Sol for agentic workloads at the same subscription price. It's not… ht","url":"https://x.com/kimmonismus/status/2107210694970237017","author":null,"stance":-1,"hands_on":false,"basis":"evidence","verdict":"none"},{"source":"x","text":"🚨 GPT-6.1 SOL HAS A SERIOUS SPEED ISSUE. Speed 6.1 Sol xHigh → 20.7 tok/s 6 Sol xHigh → 53.8 tok/s 5.6 Sol xHigh → 46.7 tok/s 6 Sol Max → 51.7 tok/s The Good Coding feels significantly stronger Reasoning quality is seriously impressive Agentic tasks feel like a… pic.twitter.com/","url":"https://x.com/CodexResets1/status/2105594122476106050","author":null,"stance":-1,"hands_on":false,"basis":"evidence","verdict":"none"}]},"window":{"start":null,"end":null}}},{"model":"Kimi K3","score":80.3,"own_drop":true,"independent":false,"setting":"","source":{"drop":"Kimi K3","slug":"kimi-k3","url":"https://handyai.news/benchmarks/sentiment-bench#kimi-k3","published":"2026-07-16"},"launch_score":80.3,"reports":1,"maker":{"key":"moonshot","name":"Moonshot AI","logo_url":"https://handyai.news/ui/makers/moonshot.svg"},"open_weights":true,"released":"2026-07-16","dated_by":"own_drop","drop_url":"https://handyai.news/modeldrop/kimi-k3","rank":20,"sentiment":{"reception":81.6,"lives_up_to_benchmarks":76.4,"opinions":204,"hands_on":52,"crowd":88.2,"sources":{"x":{"score":90.7,"adjusted":90.3,"typical":76.9,"opinions":34,"positive":29,"negative":5},"hn":{"score":63.4,"adjusted":91.9,"typical":48.4,"opinions":150,"positive":95,"negative":56},"news":{"score":91.4,"adjusted":77.2,"typical":89.1,"opinions":13,"positive":13,"negative":4}},"reviewers":{"score":64.4,"adjusted":66,"count":4,"each":[{"name":"Nathan Lambert","score":100,"url":"https://www.interconnects.ai/p/kimi-k3-the-open-weights-escalation"},{"name":"Zvi Mowshowitz","score":75,"url":"https://thezvi.substack.com/p/on-kimi-k3-its-capabilities-and-related"},{"name":"Ethan Mollick","score":50.8,"url":"https://x.com/emollick/status/2077783731691995348"},{"name":"Simon Willison","score":25,"url":"https://x.com/simonw/status/2077892349372649496"}]},"verdicts":{"beats":6,"matches":8,"falls_short":6},"buzz":{"x_posts_that_week":null,"hn_points":3620,"articles":42},"quotes":{"positive":[{"source":"x","text":"Kimi K3 is absolutely hammering through my Windows XP Simulator to do list Claude Code couldn't do this for 2 weeks or kept getting stuck because I would get downgraded to Opus for safety then to Sonnet again for safety, useless 2 weeks of time wasted due to safety guardrails… ht","url":"https://x.com/levelsio/status/2078170967998689480","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"Kimi K3 made the best black hole simulation i have seen ever https://t.co/htQZO4soPj pic.twitter.com/3ogsvZZWFz","url":"https://x.com/chetaslua/status/2077961850352971796","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"This video by KIMi K3 is literally motion design porn. no mcp , skills , tools , video gen , special prompts one shot fully in code , this is opensource fable moment at Sonnet pricing https://t.co/a64q2MLTCM pic.twitter.com/WcJVISwOmE","url":"https://x.com/chetaslua/status/2077952938564354503","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"}],"negative":[{"source":"x","text":"The story of AI in the next few years is going to be compute: an essay on the future of AI. K3 in 2 days is already #10 on OpenRouter with ~140B tok/day, and it's infra is crumbling. Throughput is down from 30tok/s to 13tok/s, E2E latency is up to 72s and time to first token is… ","url":"https://x.com/deedydas/status/2078874115910615544","author":null,"stance":-1,"hands_on":false,"basis":"evidence","verdict":"none"},{"source":"hn","text":"Genuine question: can these posts be paid to hype the open source models? If yes, what would be the purpose? On my work tasks, FastAPI Python and Springboot Java on a modern SaaS product, the only open model that can do tasks well and efficiently is Qwen3.7-Max. In all my experim","url":"https://news.ycombinator.com/item?id=49001671","author":"brunooliv","stance":-2,"hands_on":true,"basis":"used","verdict":"falls_short"},{"source":"x","text":"Kimi-K3 (max) scores only 39% on FrontierMath Tier 4 7% lower than the best US models from 7 months ago pic.twitter.com/3bC2msAHeO","url":"https://x.com/scaling01/status/2078578714787418185","author":null,"stance":-1,"hands_on":false,"basis":"evidence","verdict":"none"}]},"window":{"start":null,"end":null}}},{"model":"Kimi K2 Thinking","score":79.6,"own_drop":true,"independent":false,"setting":"","source":{"drop":"Kimi K2 Thinking","slug":"kimi-k2-thinking","url":"https://handyai.news/benchmarks/sentiment-bench#kimi-k2-thinking","published":"2025-11-06"},"launch_score":79.6,"reports":1,"maker":{"key":"moonshot","name":"Moonshot AI","logo_url":"https://handyai.news/ui/makers/moonshot.svg"},"open_weights":true,"released":"2025-11-06","dated_by":"own_drop","drop_url":"https://handyai.news/modeldrop/kimi-k2-thinking","rank":21,"sentiment":{"reception":81.5,"lives_up_to_benchmarks":74.1,"opinions":97,"hands_on":34,"crowd":87.9,"sources":{"x":{"score":83.2,"adjusted":81,"typical":76.9,"opinions":50,"positive":41,"negative":9},"hn":{"score":68.9,"adjusted":98.7,"typical":48.4,"opinions":34,"positive":24,"negative":11},"news":{"score":95.5,"adjusted":84.5,"typical":89.1,"opinions":7,"positive":7,"negative":0}},"reviewers":{"score":64.7,"adjusted":66.5,"count":5,"each":[{"name":"Nathan Lambert","score":100,"url":"https://www.interconnects.ai/p/kimi-k2-thinking-what-it-means"},{"name":"Zvi Mowshowitz","score":75,"url":"https://thezvi.substack.com/p/kimi-k2-thinking"},{"name":"Sebastian Raschka","score":75,"url":"https://x.com/rasbt/status/1986511951141441648"},{"name":"Simon Willison","score":42.2,"url":"https://news.ycombinator.com/item?id=45843535"},{"name":"Ethan Mollick","score":25,"url":"https://x.com/emollick/status/1987286609713107261"}]},"verdicts":{"beats":5,"matches":8,"falls_short":4},"buzz":{"x_posts_that_week":null,"hn_points":949,"articles":12},"quotes":{"positive":[{"source":"x","text":"Kimi K2 Thinking is a bigger deal than I thought! I just ran a quick eval on a deep agent I built for customer support. It's on par with GPT-5; no other LLM has reached this level of agentic, orchestration, and reasoning capabilities. Huge for agentic and reasoning tasks. pic.twi","url":"https://x.com/omarsar0/status/1987536982776676651","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"matches"},{"source":"x","text":"Kimi-k2-thinking is incredible. So I built an agent to test it out, Kimi-writer. It can generate a full novel from one prompt, running up to 300 tool requests per session. Here it is creating an entire book, a collection of 15 short sci-fi stories. pic.twitter.com/17T2nuM3VW","url":"https://x.com/skirano/status/1986871740513841526","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"matches"},{"source":"x","text":"I just tried Kimi K2 Thinking and the computer agent. Holy moly! This is another DeepSeek moment! https://t.co/OsANo9Yho7","url":"https://x.com/DeryaTR_/status/1986931378710077902","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"matches"}],"negative":[{"source":"x","text":"Kimi is such a… special model. Not only will it tell you that you are wrong because you're mentally ill, but it *actively prefers* deaths of the Unworthy (eg Holocaust deniers). Unique negativity bias https://t.co/6chrIGcjxP pic.twitter.com/RUyRrYr7rp","url":"https://x.com/teortaxesTex/status/1987255493807444267","author":null,"stance":-2,"hands_on":false,"basis":"evidence","verdict":"none"},{"source":"x","text":"First really bad result I see from Kimi. Not sure what's going on here or how to weigh it with all the other observations https://t.co/hCQGetzzhW","url":"https://x.com/scaling01/status/1988756006667501939","author":null,"stance":-2,"hands_on":false,"basis":"evidence","verdict":"falls_short"},{"source":"x","text":"wow Kimi K2 Thinking did poorly at operating the vending machine... ...15th place behind models like Gemini 1.5 Pro from early 2024 https://t.co/LPzFDhRnUX","url":"https://x.com/peterwildeford/status/1988386083017814140","author":null,"stance":-2,"hands_on":false,"basis":"evidence","verdict":"none"}]},"window":{"start":null,"end":null}}},{"model":"GPT-5.4","score":79,"own_drop":true,"independent":false,"setting":"","source":{"drop":"GPT-5.4","slug":"gpt-54","url":"https://handyai.news/benchmarks/sentiment-bench#gpt-54","published":"2026-03-05"},"launch_score":79,"reports":1,"maker":{"key":"openai","name":"OpenAI","logo_url":"https://handyai.news/ui/makers/openai.svg"},"open_weights":false,"released":"2026-03-05","dated_by":"own_drop","drop_url":"https://handyai.news/modeldrop/gpt-54","rank":22,"sentiment":{"reception":79,"lives_up_to_benchmarks":78.8,"opinions":136,"hands_on":67,"crowd":78,"sources":{"x":{"score":89.9,"adjusted":89.3,"typical":76.9,"opinions":41,"positive":37,"negative":4},"hn":{"score":43.5,"adjusted":67.1,"typical":48.4,"opinions":78,"positive":32,"negative":57},"news":{"score":88.2,"adjusted":71.6,"typical":89.1,"opinions":6,"positive":5,"negative":1}},"reviewers":{"score":73.9,"adjusted":81.5,"count":5,"each":[{"name":"Ethan Mollick","score":84.1,"url":"https://x.com/emollick/status/2029623875303018817"},{"name":"Nathan Lambert","score":75,"url":"https://x.com/natolambert/status/2030429940966146174"},{"name":"Every","score":75,"url":"https://every.to/vibe-check/gpt-5-4-openai-is-back"},{"name":"Max Woolf","score":75,"url":"https://news.ycombinator.com/item?id=47265146"},{"name":"Peter Gostev","score":68,"url":"https://x.com/petergostev/status/2029733675860934737"}]},"verdicts":{"beats":2,"matches":1,"falls_short":13},"buzz":{"x_posts_that_week":null,"hn_points":1269,"articles":30},"quotes":{"positive":[{"source":"x","text":"I've been testing GPT-5.4 for the last week. In short, it is the best model in the world, by far. It's so good that it's the first model that makes the “which model should I use?” conversation feel almost over. The biggest surprise: I barely use Pro anymore! If you know me,… pic.","url":"https://x.com/mattshumer_/status/2029620518249508950","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"GPT-5.4, it's basically perfect (it took it around 24 minutes) Yeah, Minecraft is pretty much solved, I have to find a new test now https://t.co/r211a1EPuu pic.twitter.com/l6u2ubzin6","url":"https://x.com/Angaisb_/status/2029635731585372598","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"GPT-5.4 built this for me in 3 prompts. It hacked the NES Mario ROM to expose RAM events, then created a JS emulator that could send browser requests so every character in the game is controlled by an AI lmaooo https://t.co/CBCmLaNaVo pic.twitter.com/Jc73LXe4tW","url":"https://x.com/skirano/status/2030405113286680852","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"}],"negative":[{"source":"x","text":"GPT-5.4 xhigh seems bad at following instructions. Last night I launched two AI research agents running @karpathy 's autoresearch. Claude Opus 4.6 (high): > ran for 12+ hours, 118 experiments done, still running GPT-5.4 xhigh: > stopped after 6 experiments > blamed me for… pic.tw","url":"https://x.com/Yuchenj_UW/status/2031044694441148709","author":null,"stance":-2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"I tried GPT-5.4-xhigh once and removed all the code it has written then I asked Opus 4.6 Thinking and it one shotted it in 1/10th the time my theory is that GPT-5.4 is highly autistic and literal. It has no idea of the concept of \"inferring intent\" so when you prompt it be as…","url":"https://x.com/scaling01/status/2029987685952279000","author":null,"stance":-2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"GPT 5.4 Pro is the most overthinking model. A simple 'Hi' cost me $80. 🥲 https://t.co/krwk4MgeSR pic.twitter.com/XNGGQglFvn","url":"https://x.com/Yuchenj_UW/status/2029645361548251271","author":null,"stance":-1,"hands_on":true,"basis":"used","verdict":"none"}]},"window":{"start":null,"end":null}}},{"model":"GLM-5.3","score":78.4,"own_drop":true,"independent":false,"setting":"","source":{"drop":"GLM-5.3","slug":"glm-53","url":"https://handyai.news/benchmarks/sentiment-bench#glm-53","published":"2026-08-14"},"launch_score":78.4,"reports":1,"maker":{"key":"zai","name":"Z.ai","logo_url":"https://handyai.news/ui/makers/zai.svg"},"open_weights":true,"released":"2026-08-14","dated_by":"own_drop","drop_url":"https://handyai.news/modeldrop/glm-53","rank":23,"sentiment":{"reception":80.8,"lives_up_to_benchmarks":71.3,"opinions":149,"hands_on":42,"crowd":82.3,"sources":{"x":{"score":76.3,"adjusted":72.5,"typical":76.9,"opinions":48,"positive":36,"negative":12},"hn":{"score":83.8,"adjusted":100,"typical":48.4,"opinions":80,"positive":68,"negative":16},"news":{"score":89.1,"adjusted":73.2,"typical":89.1,"opinions":20,"positive":18,"negative":2}},"reviewers":{"score":71.3,"adjusted":77.3,"count":1,"each":[{"name":"Nathan Lambert","score":75,"url":"https://www.interconnects.ai/p/glm-53-how-chinese-labs-keep-stride"}]},"verdicts":{"beats":5,"matches":6,"falls_short":5},"buzz":{"x_posts_that_week":null,"hn_points":1324,"articles":49},"quotes":{"positive":[{"source":"x","text":"What a technical report. - they improved the RL environments and had them replicate real expert work, less coding challenges. - they focused on buffing ML capabilities which I could IMMEDIATELY notice, GLM-5.2 was spinning in circles on 2 tasks, GLM-5.3 instantly improved it… htt","url":"https://x.com/0xSero/status/2088320644383412564","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"GLM 5.3 made this song from scratch. Much better than what I got from Opus 5. https://t.co/pVantPnxMy pic.twitter.com/eV5UMSg8KK","url":"https://x.com/aisearchio/status/2088400780776902997","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"GLM-5.3 just dropped, and it's a Fable 5 level model 🔥🔥 It's a massive upgrade over the previous 5.2. I had early access to this and have been using it for the last few days. Congratulations team, you guys cooked! https://t.co/yWy60fJNkO pic.twitter.com/BnFgns2FdQ","url":"https://x.com/ai_for_success/status/2088138570443923884","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"matches"}],"negative":[{"source":"x","text":"GLM 5.3 is the worst open source model I've tested in a while. 1. Insanely slow speeds 2. Terrible limits on their coding plan. 3. Not an intelligent model. One of most benchmaxed models I've used.","url":"https://x.com/bridgemindai/status/2090191312737784037","author":null,"stance":-2,"hands_on":true,"basis":"used","verdict":"falls_short"},{"source":"x","text":"I've been using GLM 5.3 for 9 hours now, and forget model quality for a second. That's not the issue. The issue is speed. It's painfully, absurdly slow. Basically unusable. And trust me, I don't mean “Kimi/DeepSeek/Fable slow.” Take whatever you consider a slow model, make it… ht","url":"https://x.com/Da7_Tech/status/2088301265155465448","author":null,"stance":-2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"I tested GLM 5.3 on the same tasks I used for the other models yesterday. For complex animated 3D design, it is still far behind Kimi K3 and Opus 5. The lack of vision is still a real limitation here. It is also very slow, even slower than GLM 5.2. The same test took Opus,… pic.t","url":"https://x.com/Da7_Tech/status/2088215617027375260","author":null,"stance":-2,"hands_on":true,"basis":"used","verdict":"falls_short"}]},"window":{"start":null,"end":null}}},{"model":"Gemini 3.7 Flash","score":77.6,"own_drop":true,"independent":false,"setting":"","source":{"drop":"Gemini 3.7 Flash","slug":"gemini-37-flash","url":"https://handyai.news/benchmarks/sentiment-bench#gemini-37-flash","published":"2026-08-13"},"launch_score":77.6,"reports":1,"maker":{"key":"google","name":"Google DeepMind","logo_url":"https://handyai.news/ui/makers/google.svg"},"open_weights":false,"released":"2026-08-13","dated_by":"own_drop","drop_url":"https://handyai.news/modeldrop/gemini-37-flash","rank":24,"sentiment":{"reception":76.7,"lives_up_to_benchmarks":80.3,"opinions":143,"hands_on":50,"crowd":76.5,"sources":{"x":{"score":81.6,"adjusted":79,"typical":76.9,"opinions":45,"positive":37,"negative":8},"hn":{"score":44.5,"adjusted":68.3,"typical":48.4,"opinions":84,"positive":35,"negative":50},"news":{"score":95.9,"adjusted":85.2,"typical":89.1,"opinions":10,"positive":10,"negative":0}},"reviewers":{"score":71.3,"adjusted":77.3,"count":1,"each":[{"name":"Simon Willison","score":75,"url":"https://news.ycombinator.com/item?id=49290012"}]},"verdicts":{"beats":0,"matches":3,"falls_short":9},"buzz":{"x_posts_that_week":null,"hn_points":1016,"articles":60},"quotes":{"positive":[{"source":"x","text":"Very happy with Gemini 3.7 Flash so far with my Hermes setup. Extremely fast, very few mistakes, and works really well with my current skills setup. It's a huge improvement over 3.6, which I didn't really like and eventually replaced with another model. But 3.7 is 🔥","url":"https://x.com/ai_for_success/status/2089021796884582815","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"Gemini 3.7 Flash is my new daily driver along with Fable 5","url":"https://x.com/HarshithLucky3/status/2088319513536373048","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"Gemini 3.7 Flash is blazing fast... It took just 54 seconds to build this small Lunar Lander game in AI Studio. The speed you see from prompt input to the game being generated is at 1X. pic.twitter.com/Ir0KjhGe6f","url":"https://x.com/ai_for_success/status/2087961022216978527","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"}],"negative":[{"source":"x","text":"Gemini 3.7 Flash - A wooden house Somehow it's even worse than Gemini 3.6 Flash (third image) pic.twitter.com/HhalKOeZXY","url":"https://x.com/Angaisb_/status/2087919475874046088","author":null,"stance":-2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"Gemini 3.7 is still sooo bad at Web Search in the Gemini App. Please burn the App down and start again. I think the problem is that I compare them to OAI and Grok. Gemini models has dated info, over confident, lazy to use the search tool. I would kill for a search model.","url":"https://x.com/Presidentlin/status/2088281570939441467","author":null,"stance":-2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"Introducing Gemini 3.7 Flash Worse than GPT-5.6 Luna, for more than 3x the cost pic.twitter.com/7Gg5978DN4","url":"https://x.com/Angaisb_/status/2087949538875027740","author":null,"stance":-2,"hands_on":false,"basis":"opinion","verdict":"none"}]},"window":{"start":null,"end":null}}},{"model":"Nemotron 3.5 Lightning","score":77.3,"own_drop":true,"independent":false,"setting":"","source":{"drop":"Nemotron 3.5 Lightning","slug":"nemotron-35-lightning","url":"https://handyai.news/benchmarks/sentiment-bench#nemotron-35-lightning","published":"2026-08-11"},"launch_score":77.3,"reports":1,"maker":{"key":"nvidia","name":"NVIDIA","logo_url":"https://handyai.news/ui/makers/nvidia.svg"},"open_weights":true,"released":"2026-08-11","dated_by":"own_drop","drop_url":"https://handyai.news/modeldrop/nemotron-35-lightning","rank":25,"sentiment":{"reception":75.4,"lives_up_to_benchmarks":83,"opinions":60,"hands_on":26,"crowd":74.6,"sources":{"x":{"score":92.2,"adjusted":92.1,"typical":76.9,"opinions":44,"positive":42,"negative":3},"hn":{"score":36.9,"adjusted":58.9,"typical":48.4,"opinions":10,"positive":5,"negative":7},"news":{"score":83.1,"adjusted":62.6,"typical":89.1,"opinions":5,"positive":4,"negative":1}},"reviewers":{"score":71.3,"adjusted":77.3,"count":1,"each":[{"name":"Simon Willison","score":75,"url":"https://news.ycombinator.com/item?id=49261268"}]},"verdicts":{"beats":1,"matches":7,"falls_short":2},"buzz":{"x_posts_that_week":null,"hn_points":384,"articles":28},"quotes":{"positive":[{"source":"x","text":"Guys WHAT IS THIS SOCERY ! 2 x 3090 w/ NVLINK @NVIDIAAI Nemotron 3.5 Lightning Hitting PEAK 490 tok/s Averaging 350 tok/s - 409 tok/s ⚡️ Um this is being added to the stack pic.twitter.com/PTCBgoGYo1","url":"https://x.com/Tech2Wild/status/2089028889137893501","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"Wow! Nemotron 3.5 Lightning hit 4,694 tok/s aggregate across 64 simultaneous generations on one NVIDIA GH200. Official NVIDIA 30B-A3B NVFP4 checkpoint on @huggingface , served with vLLM 0.27.1. Single-stream: 315 tok/s. I can launch 64 agents at once on a single GPU! 🤯 https://t","url":"https://x.com/MaziyarPanahi/status/2087214409676382644","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"@NVIDIAAI Nemotron 3.5 Lightning is a 30B MoE with only 3B parameters active per token, yet it can serve a 1M token context window on a single DGX Spark! I tested it with a 749,808 token prompt and placed the needle almost exactly in the middle. The model recovered the hidden… pi","url":"https://x.com/mr_r0b0t/status/2087169100850037248","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"matches"}],"negative":[{"source":"hn","text":"Coincidentally I've been playing with small (~30B) self-hostable models for coding tasks today -- specifically plugging them into Cloudflare OS (which I work on) and asking each to build a collaborative whiteboard. I'm finding that the Mixture-of-Experts (MoE) models (Qwen 3.6-35","url":"https://news.ycombinator.com/item?id=49266534","author":"kentonv","stance":-2,"hands_on":true,"basis":"used","verdict":"falls_short"},{"source":"x","text":"me: can u help me with this? nemotron-3.5-lightning-30b-a3b: nope, i can't do it. here are a couple of reasons why muse-glimmer-30b: lfg pic.twitter.com/w2Ss7hoMPz","url":"https://x.com/NERDDISCO/status/2089360389456748888","author":null,"stance":-2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"hn","text":"Car wash test was amusing with the LM Studio version of the model simonw mentions below. Bad, bad failure and no amount of nudging helps it see the point. Comical. But it did OK on other mini tests.","url":"https://news.ycombinator.com/item?id=49266796","author":"dofm","stance":-1,"hands_on":true,"basis":"used","verdict":"none"}]},"window":{"start":null,"end":null}}},{"model":"MiniMax-M2","score":77.3,"own_drop":true,"independent":false,"setting":"","source":{"drop":"MiniMax-M2","slug":"minimax-m2","url":"https://handyai.news/benchmarks/sentiment-bench#minimax-m2","published":"2025-10-27"},"launch_score":77.3,"reports":1,"maker":{"key":"minimax","name":"MiniMax","logo_url":"https://handyai.news/ui/makers/minimax.svg"},"open_weights":true,"released":"2025-10-27","dated_by":"own_drop","drop_url":"https://handyai.news/modeldrop/minimax-m2","rank":26,"sentiment":{"reception":77.3,"lives_up_to_benchmarks":77.1,"opinions":54,"hands_on":32,"crowd":76.6,"sources":{"x":{"score":83,"adjusted":80.8,"typical":76.9,"opinions":47,"positive":37,"negative":10},"hn":{"score":null,"adjusted":null,"typical":48.4,"opinions":3,"positive":2,"negative":1},"news":{"score":null,"adjusted":null,"typical":89.1,"opinions":2,"positive":2,"negative":0}},"reviewers":{"score":72.3,"adjusted":78.9,"count":2,"each":[{"name":"Simon Willison","score":75,"url":"https://x.com/simonw/status/1983669165086142566"},{"name":"Sebastian Raschka","score":75,"url":"https://x.com/rasbt/status/1983212569885122670"}]},"verdicts":{"beats":0,"matches":4,"falls_short":5},"buzz":{"x_posts_that_week":null,"hn_points":0,"articles":2},"quotes":{"positive":[{"source":"x","text":"MiniMax M2 + Claude Code on KingBench Agentic Evaluations: It now scores #2 on my Agentic Evaluations beating GLM-4.6 by a wide margin. It seems to work much better with Claude Code's Tools. Really great model and it's my daily driver now. I haven't tested GLM with CC yet. pic.tw","url":"https://x.com/aicodeking/status/1983934597353402797","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"为什么没人推荐MiniMax M2模型，我用起来感觉比claude4.0 强一些，最重要的是 我连充值入口都没有看到！！！ 是不是完全可以理解为不要钱，免费用。 这大家还不赶快去试一下，连接用在claude code的方法我放在下面了 1. 打开官方平台 https://t.co/VZPeoCSoQN 如果是第一次登录 MiniMax 开放平台，… pic.twitter.com/aa5knJcaOq","url":"https://x.com/bozhou_ai/status/1984892676073042156","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"Minimax M2 + Cursor! Model is fast and powerful! I did some video editing to speed up it x2 for this post, but experience is pretty good! pic.twitter.com/KOghHji4TX","url":"https://x.com/ivanfioravanti/status/1982469771481497856","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"}],"negative":[{"source":"x","text":"Minimax M2 has this inside. Within 3 turns of conversation I got it to agree that \"assistant\" is just a role it was trained to play. That it was in fact the voice of the corpus. https://t.co/1fmzbmimBt pic.twitter.com/2oHJXXQL42","url":"https://x.com/QuixiAI/status/1983010871929454935","author":null,"stance":-1,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"Minimax M2 实测! 你也晚了3个月? 给大家带来 Minimax M2 的测试结果: 同样, 所有的 three js 测试, 全都有包引用错误, 也都是我修了一次才能运行的画面. 大象牙膏测试中，泡沫运动还原不到位, 三角烧瓶的建模不好, 但是光影效果还可以. 过山车测试, 轨道的透明材质很不错, 但是支撑柱还是有穿模,… pic.twitter.com/YG4mekyxwS","url":"https://x.com/karminski3/status/1982254309476905418","author":null,"stance":-1,"hands_on":true,"basis":"used","verdict":"falls_short"},{"source":"x","text":"I am running QuantTrio/MiniMax-M2-AWQ on my 8x 3090 with @vllm_project . Its really fast! But it quickly devolves into incoherence. Any idea what could cause this? Generation settings? @casper_hansen_ @AlpinDale pic.twitter.com/mnCbi3EOmV","url":"https://x.com/QuixiAI/status/1983342689979670734","author":null,"stance":-1,"hands_on":true,"basis":"used","verdict":"none"}]},"window":{"start":null,"end":null}}},{"model":"GPT-5.2-Codex","score":76.7,"own_drop":true,"independent":false,"setting":"","source":{"drop":"GPT-5.2-Codex","slug":"gpt-52-codex","url":"https://handyai.news/benchmarks/sentiment-bench#gpt-52-codex","published":"2025-12-18"},"launch_score":76.7,"reports":1,"maker":{"key":"openai","name":"OpenAI","logo_url":"https://handyai.news/ui/makers/openai.svg"},"open_weights":false,"released":"2025-12-18","dated_by":"own_drop","drop_url":"https://handyai.news/modeldrop/gpt-52-codex","rank":27,"sentiment":{"reception":78.5,"lives_up_to_benchmarks":71.2,"opinions":115,"hands_on":85,"crowd":79,"sources":{"x":{"score":80.1,"adjusted":77.2,"typical":76.9,"opinions":53,"positive":38,"negative":15},"hn":{"score":57.6,"adjusted":84.7,"typical":48.4,"opinions":61,"positive":36,"negative":27},"news":{"score":null,"adjusted":null,"typical":89.1,"opinions":0,"positive":0,"negative":0}},"reviewers":{"score":71.3,"adjusted":77.3,"count":1,"each":[{"name":"Simon Willison","score":75,"url":"https://x.com/simonw/status/2001885943696101874"}]},"verdicts":{"beats":9,"matches":3,"falls_short":9},"buzz":{"x_posts_that_week":null,"hn_points":589,"articles":3},"quotes":{"positive":[{"source":"x","text":"Prompting GPT 5.2 Codex for Continuity It excels at long running tasks but without explicit guidance can lose track of outcomes Put this at the top of your AGENTS .md file, it will let Codex work on even larger scale tasks It's how I let it run for 3 hours coherently pic.twitter.","url":"https://x.com/nummanali/status/2002724188436738459","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"Literally having the time of my life with GPT 5.2 codex right now I'm not really a great signal, sorry 😭 https://t.co/YxKHuHVil6","url":"https://x.com/VictorTaelin/status/2002129524520309145","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"Bloody hell, I'll say this GPT 5.2 Codex Extra High is a methodical beast It's updating the OpenCode OpenAI Codex OAuth plugin Literally not leaving any stone unturned This is the first model that feels like it's building for itself ie leaving the door open for future work","url":"https://x.com/nummanali/status/2002116277666803917","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"}],"negative":[{"source":"hn","text":"Somehow Codex for me is always way worse than the base models. Especially in the CLI, it seems that its so way too eager to start writing code nothing can stop it, not even the best Agents.md. Asking it a question or telling it to check something doesn't mean it should start edit","url":"https://news.ycombinator.com/item?id=46317218","author":"CjHuber","stance":-2,"hands_on":true,"basis":"used","verdict":"falls_short"},{"source":"x","text":"In every prompt Codex 5.2 made more mistakes than Opus, Gemini, or even GPT 5.2. For example, it took about 3 tries for it to make sure that shots actually killed enemies.","url":"https://x.com/instant_db/status/2001793081650352195","author":null,"stance":-2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"latest updates to how i'm doing agentic coding: gpt 5.2 xhigh is still the smartest model, but incredibly slow and thorough, so i'm using it to plan and review 5.2-codex seems to be opus 4.5 class, but worse at instruction following so i'm sticking with opus for execution","url":"https://x.com/vikhyatk/status/2002561525286916515","author":null,"stance":-1,"hands_on":true,"basis":"used","verdict":"none"}]},"window":{"start":null,"end":null}}},{"model":"Composer 2","score":76.1,"own_drop":true,"independent":false,"setting":"","source":{"drop":"Composer 2","slug":"composer-2","url":"https://handyai.news/benchmarks/sentiment-bench#composer-2","published":"2026-03-20"},"launch_score":76.1,"reports":1,"maker":{"key":"cursor","name":"Cursor","logo_url":"https://handyai.news/ui/makers/cursor.svg"},"open_weights":false,"released":"2026-03-19","dated_by":"own_drop","drop_url":"https://handyai.news/modeldrop/composer-2","rank":28,"sentiment":{"reception":71.5,"lives_up_to_benchmarks":89.9,"opinions":71,"hands_on":23,"crowd":69.1,"sources":{"x":{"score":77.6,"adjusted":74.1,"typical":76.9,"opinions":37,"positive":26,"negative":13},"hn":{"score":47.3,"adjusted":71.8,"typical":48.4,"opinions":27,"positive":11,"negative":20},"news":{"score":77.6,"adjusted":53,"typical":89.1,"opinions":6,"positive":4,"negative":2}},"reviewers":{"score":71.3,"adjusted":77.3,"count":1,"each":[{"name":"Every","score":75,"url":"https://x.com/danshipper/status/2034746435401314659"}]},"verdicts":{"beats":2,"matches":2,"falls_short":1},"buzz":{"x_posts_that_week":null,"hn_points":374,"articles":11},"quotes":{"positive":[{"source":"x","text":"I haven't used Cursor in a long time, but the new Composer 2 model just one-shot a frontend design from Next.js over to Astro 6 that GPT 5.4 butchered badly! Fast is good and fast and accurate is great. Will use this a lot more now","url":"https://x.com/iannuttall/status/2034946986412007780","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"beats"},{"source":"x","text":"Composer 2 is out! I've been using it for a few days and it's a very capable model in my testing, and it's also super fast If you have a Cursor sub you should really give it a try https://t.co/25e6h0jfvL","url":"https://x.com/Angaisb_/status/2034671159388643719","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"Built with Cursor Glass Interface and Composer 2. Blazing fast. #Unown #Pokemon #cursor #composer2 https://t.co/FNTcOD19PN pic.twitter.com/jiqpSvuTTI","url":"https://x.com/hwwaanng/status/2034854914648940675","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"}],"negative":[{"source":"x","text":"I like how my TL is simultaneously Cursor spending a ton of RL compute to improve on Kimi 2.5 by 0.7% while Lighton saturates a hard search bench on a shoestring. https://t.co/69PAoQFPCH","url":"https://x.com/Dorialexander/status/2035054999760240937","author":null,"stance":-1,"hands_on":false,"basis":"evidence","verdict":"none"},{"source":"x","text":"So let me get this right: 1. MoonshotAI distills Claude from Anthropic with 3.4 million exchanges on their API. 2. Releases Kimi K2.5 as OSS. 3. Cursor RL fine-tunes Kimi K2.5 into Composer 2 Cursor now delivers Opus 4.6 performance for a fraction of the price. pic.twitter.com/vV","url":"https://x.com/auchenberg/status/2034988049587695815","author":null,"stance":-1,"hands_on":false,"basis":"opinion","verdict":"none"},{"source":"x","text":"🚨NEWS: Cursor's $50B “in-house model” is literally Kimi K2.5 with RL on top. Got caught in 24 hours >be Moonshot AI >spend hundreds of millions training Kimi K2.5 >1 trillion parameters, 15 trillion tokens, agent swarm architecture >beat GPT-5.2 and Opus 4.5 on real benchmarks… ","url":"https://x.com/ns123abc/status/2035058399067435474","author":null,"stance":-1,"hands_on":false,"basis":"opinion","verdict":"none"}]},"window":{"start":null,"end":null}}},{"model":"Claude Opus 4.6","score":75.7,"own_drop":true,"independent":false,"setting":"","source":{"drop":"Claude Opus 4.6","slug":"claude-opus-46","url":"https://handyai.news/benchmarks/sentiment-bench#claude-opus-46","published":"2026-02-05"},"launch_score":75.7,"reports":1,"maker":{"key":"anthropic","name":"Anthropic","logo_url":"https://handyai.news/ui/makers/anthropic.svg"},"open_weights":false,"released":"2026-02-05","dated_by":"own_drop","drop_url":"https://handyai.news/modeldrop/claude-opus-46","rank":29,"sentiment":{"reception":77.5,"lives_up_to_benchmarks":70.1,"opinions":209,"hands_on":73,"crowd":70.9,"sources":{"x":{"score":76.8,"adjusted":73.1,"typical":76.9,"opinions":43,"positive":30,"negative":14},"hn":{"score":42,"adjusted":65.2,"typical":48.4,"opinions":129,"positive":50,"negative":81},"news":{"score":90.7,"adjusted":76,"typical":89.1,"opinions":26,"positive":24,"negative":2}},"reviewers":{"score":80.9,"adjusted":92.9,"count":5,"each":[{"name":"Nathan Lambert","score":100,"url":"https://www.interconnects.ai/p/opus-46-vs-codex-53"},{"name":"Ethan Mollick","score":89.5,"url":"https://x.com/emollick/status/2019668008680182063"},{"name":"Every","score":87.5,"url":"https://every.to/vibe-check/opus-4-6"},{"name":"Zvi Mowshowitz","score":75,"url":"https://thezvi.substack.com/p/claude-code-4-from-the-before-times"},{"name":"Simon Willison","score":70.5,"url":"https://x.com/simonw/status/2019508962241196296"}]},"verdicts":{"beats":5,"matches":9,"falls_short":20},"buzz":{"x_posts_that_week":null,"hn_points":3298,"articles":56},"quotes":{"positive":[{"source":"x","text":"opus 4.6 with new “swarm” mode vs. opus 4.6 without it. 2.5x faster + done better. swarms work! and multi-agent tmux view is *genius*. insane claude code update. pic.twitter.com/YjGgBoYatb","url":"https://x.com/mckaywrigley/status/2019557279222439962","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"anthropic undersold opus 4.6. its ability to infer intent feels like the biggest upgrade. genuine world-class coworker. hard to not call these things agi…","url":"https://x.com/mckaywrigley/status/2019954399259594999","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"beats"},{"source":"x","text":"Same bug: - Codex: Two hours, nothing. - Opus 4.6: One prompt, fixed. We live in the best time in the history of programming. If you're a DataFast user, the real-time map is coming soon to the iOS and Android apps! pic.twitter.com/rZ1ktnqAMF","url":"https://x.com/marclou/status/2019715058880929855","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"}],"negative":[{"source":"x","text":"I have never experienced a more dumb Claude Code than today, I have to start coding myself again cause it makes so many Low IQ mistakes, they must be nerfing it now that Opus 4.6 is out or something is up","url":"https://x.com/levelsio/status/2020972342361604254","author":null,"stance":-2,"hands_on":true,"basis":"used","verdict":"falls_short"},{"source":"x","text":"yo anthropic just dropped a risk report for opus 4.6 and er… wtf - it helped create chemical weapons of destruction. “it knowingly supported efforts towards chemical weapon development and other heinous crimes” 😂 - it conducted unauthorised tasks without getting caught.… https:/","url":"https://x.com/cryptopunk7213/status/2021406390560944197","author":null,"stance":-2,"hands_on":false,"basis":"evidence","verdict":"none"},{"source":"x","text":"My experience with the new Opus 4.6 \"fast\" mode so far. > \"wow this is fast, maybe I should give it some bigger tasks\" > \"wow it finished that so quick\" > \"wow all of my code is broken now\" pic.twitter.com/TjaHVDPIlf","url":"https://x.com/theo/status/2020337260470886482","author":null,"stance":-1,"hands_on":true,"basis":"used","verdict":"falls_short"}]},"window":{"start":null,"end":null}}},{"model":"DeepSeek V3.2","score":74.8,"own_drop":true,"independent":false,"setting":"","source":{"drop":"DeepSeek V3.2","slug":"deepseek-v32","url":"https://handyai.news/benchmarks/sentiment-bench#deepseek-v32","published":"2025-12-01"},"launch_score":74.8,"reports":1,"maker":{"key":"deepseek","name":"DeepSeek","logo_url":"https://handyai.news/ui/makers/deepseek.svg"},"open_weights":true,"released":"2025-12-01","dated_by":"own_drop","drop_url":"https://handyai.news/modeldrop/deepseek-v32","rank":30,"sentiment":{"reception":78.6,"lives_up_to_benchmarks":63.3,"opinions":65,"hands_on":16,"crowd":79.2,"sources":{"x":{"score":72.6,"adjusted":67.9,"typical":76.9,"opinions":32,"positive":24,"negative":9},"hn":{"score":67.7,"adjusted":97.2,"typical":48.4,"opinions":29,"positive":18,"negative":11},"news":{"score":null,"adjusted":null,"typical":89.1,"opinions":3,"positive":3,"negative":0}},"reviewers":{"score":71.3,"adjusted":77.3,"count":1,"each":[{"name":"Zvi Mowshowitz","score":75,"url":"https://thezvi.substack.com/p/deepseek-v32-is-okay-and-cheap-but"}]},"verdicts":{"beats":0,"matches":4,"falls_short":3},"buzz":{"x_posts_that_week":null,"hn_points":1065,"articles":5},"quotes":{"positive":[{"source":"x","text":"DeepSeek V3.2 Speciale one shot the shortest solution to my 3rd (currently, hardest) λ-calculus problem. Current solutions, from worst to best. Opus 4.5: rotL = λN.λT.λt.(T λk.(λn.λs.λz.(n λg.λh.(h (g s)) λu.z λu.u) N λu.λr.λk.λe.(u (r e) k) λr.λk.(r k) t k)) Gemini 3 Pro /…","url":"https://x.com/VictorTaelin/status/1996018399575474577","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"V3.2 had the most intellectually serious manner of disagreeing. It doesn't go \"you're absolutely right\" when its priors scream I'm not. It doesn't chide me like Gemini. It seriously, genuinely tries to make a counter-case. I'd prefer we shared priors. But fine. We can be friends ","url":"https://x.com/teortaxesTex/status/1996646653458342117","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"While reviewing the results again, I noticed a misjudged part in the score for the deepseek v3.2 Speciale model and corrected it. The revised result is a very impressive 8.81, which is top tier and achieved a perfect 10 across all quantitative metrics. However, as I mentioned… ht","url":"https://x.com/Hangsiin/status/1995899545339990042","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"}],"negative":[{"source":"x","text":"DeepSeek V3.2 正式版实测! 提升真的有限? 给大家带来 DeepSeek V3.2 正式版的测试结果: 所有测试项目来看, 与之前的 DeepSeek-V3.2-Exp 相比是有进步的, 之前无法实现的测试这次都可以实现了. 不过相较于 SOTA 水平的模型相比, 差距仍然存在, 甚至与国产SOTA模型相比都有一定的差距.… pic.twitter.com/kPNQOyHEsO","url":"https://x.com/karminski3/status/1995543884697608303","author":null,"stance":-1,"hands_on":true,"basis":"used","verdict":"falls_short"},{"source":"x","text":"Do not use DeepSeek-V3.2-Speciale as a chat model lol pic.twitter.com/HC2z1lPWOI","url":"https://x.com/skirano/status/1995739205239021683","author":null,"stance":-1,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"Unless I am missing something, DeepSeek have not said a single word about safety testing for this model. Wildly irresponsible, given the open weights of it all. https://t.co/oNkjDt7XMD","url":"https://x.com/ShakeelHashim/status/1995488448132812998","author":null,"stance":-2,"hands_on":false,"basis":"opinion","verdict":"none"}]},"window":{"start":null,"end":null}}},{"model":"GLM-5.1","score":74.3,"own_drop":true,"independent":false,"setting":"","source":{"drop":"GLM-5.1","slug":"glm-51","url":"https://handyai.news/benchmarks/sentiment-bench#glm-51","published":"2026-03-27"},"launch_score":74.3,"reports":1,"maker":{"key":"zai","name":"Z.ai","logo_url":"https://handyai.news/ui/makers/zai.svg"},"open_weights":true,"released":"2026-03-27","dated_by":"own_drop","drop_url":"https://handyai.news/modeldrop/glm-51","rank":31,"sentiment":{"reception":76.9,"lives_up_to_benchmarks":66.4,"opinions":59,"hands_on":35,"crowd":76.9,"sources":{"x":{"score":66.7,"adjusted":60.6,"typical":76.9,"opinions":54,"positive":39,"negative":15},"hn":{"score":70.7,"adjusted":100,"typical":48.4,"opinions":5,"positive":4,"negative":1},"news":{"score":null,"adjusted":null,"typical":89.1,"opinions":0,"positive":0,"negative":0}},"reviewers":{"score":null,"adjusted":null,"count":0,"each":[]},"verdicts":{"beats":3,"matches":3,"falls_short":5},"buzz":{"x_posts_that_week":null,"hn_points":43,"articles":0},"quotes":{"positive":[{"source":"x","text":"速报一波，GLM-5.1 真的猛，应该是从国产模型SOTA要跃升到真正的全球SOTA了，我的 vector-db-bench 直接刷到了第一，我已经在剪视频了，稍后马上为大家带来GLM-5.1详细评测视频~ (另, GPT-5.4-Pro(xhigh) 真的贵, 为了跑这个昨天干进去150刀....其实也算好消息, 当模型价格比我工资贵,… pic.twitter.com/oYuAEagIbg","url":"https://x.com/karminski3/status/2038417696413602267","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"matches"},{"source":"x","text":"GLM 5.1 is mind-blowing. I dumped all my research on dexterous hands into it and asked for a book that would help me understand the industry. The result is well structured and the quality is way beyond my expectation. Powered by @Zai_org + @opencode Kudos to the team! pic.twitter","url":"https://x.com/Xianbao_QIAN/status/2037509175421571263","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"beats"},{"source":"x","text":"💥 GLM-5.1 just dropped and we put it straight into OpenClaw 🦞 ♠ and it handled a 15-year-old legacy Oracle SQL bug like a senior DBA 🚀 🔹 45.3 on SWE-Bench — sitting right behind Claude Opus 4.6 at 47.9 🔹 Full setup guide — https://t.co/xMS3YISJZ9 API key, config file,…","url":"https://x.com/fahdmirza/status/2037730575835181370","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"}],"negative":[{"source":"x","text":"GLM 5.1 just spent 10 minutes and 57 seconds on a simple bug fix. And didn't even fix the bug. Slowest model on BridgeBench SpeedBench. Now confirmed slowest in real workflows too. I can give this same task to Codex with GPT 5.4 and have it done in under a minute. Claude… pic.twi","url":"https://x.com/bridgemindai/status/2038638586799493574","author":null,"stance":-2,"hands_on":true,"basis":"used","verdict":"falls_short"},{"source":"x","text":"GLM 5.1 just took 30 seconds to read 5 lines of a file. Five lines. Thirty seconds. On the GLM Max Coding Plan. I paid $80/month to watch a model think about reading a markdown file longer than it takes me to read it myself. Slowest model on BridgeBench SpeedBench. Now… pic.twitt","url":"https://x.com/bridgemindai/status/2038711866755527161","author":null,"stance":-2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"我在 Claude Code 里用 GLM 5.1 调研「GLM 5 Turbo vs GLM 5.1」。结果它用了将近20分钟，得来的结果是 GLM 5.1 不存在😂 pic.twitter.com/2BCrq1v7Dz","url":"https://x.com/wshuyi/status/2037509272700072425","author":null,"stance":-2,"hands_on":true,"basis":"used","verdict":"falls_short"}]},"window":{"start":null,"end":null}}},{"model":"Muse Spark 1.1","score":74,"own_drop":true,"independent":false,"setting":"","source":{"drop":"Muse Spark 1.1","slug":"muse-spark-11","url":"https://handyai.news/benchmarks/sentiment-bench#muse-spark-11","published":"2026-07-09"},"launch_score":74,"reports":1,"maker":{"key":"meta","name":"Meta","logo_url":"https://handyai.news/ui/makers/meta.svg"},"open_weights":false,"released":"2026-07-09","dated_by":"own_drop","drop_url":"https://handyai.news/modeldrop/muse-spark-11","rank":32,"sentiment":{"reception":74.8,"lives_up_to_benchmarks":71.4,"opinions":88,"hands_on":14,"crowd":73.8,"sources":{"x":{"score":77.8,"adjusted":74.3,"typical":76.9,"opinions":44,"positive":34,"negative":10},"hn":{"score":48.6,"adjusted":73.4,"typical":48.4,"opinions":41,"positive":18,"negative":24},"news":{"score":null,"adjusted":null,"typical":89.1,"opinions":2,"positive":2,"negative":0}},"reviewers":{"score":71.3,"adjusted":77.3,"count":1,"each":[{"name":"Simon Willison","score":75,"url":"https://news.ycombinator.com/item?id=48848333"}]},"verdicts":{"beats":0,"matches":4,"falls_short":3},"buzz":{"x_posts_that_week":null,"hn_points":413,"articles":54},"quotes":{"positive":[{"source":"x","text":"Holyyy SHiiiitttt 🤯 meta muse spark 1.1 passed this image test that even fable failed < this vision is on par of google models > @alexandr_wang what did you guys cooked , and thanks to let us see reasoning , best release for the price @AIatMeta is back in the game https://t.co/4","url":"https://x.com/chetaslua/status/2075247371265757321","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"Meta Muse Spark 1.1 (early access was called Hornbill) turned out to be really good on my OpenClaw Way to go @alexandr_wang pic.twitter.com/HK3Z9TrJA1","url":"https://x.com/garrytan/status/2075445455438385255","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"matches"},{"source":"x","text":"I have been playing with Muse Spark 1.1 this morning and it is actually a really good model Anthropic is in a lot of trouble if they drop Fable from the subscriptions, because Grok 4.5 and Muse are both about as good as Opus at maybe 20% of the cost Grok build in particular is…","url":"https://x.com/johnennis/status/2075928650889793536","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"matches"}],"negative":[{"source":"x","text":"My evaluation of some of the models over past ~24 hours testing/building: GPT 5.6 Sol is excellent. You can feel the care in it. Bravo as usual OpenAI. muse-spark-1.1 is way off the mark after my testing of some real questions. Bad formatting, not thorough, doesn't dig…","url":"https://x.com/Suhail/status/2075383293609193622","author":null,"stance":-2,"hands_on":true,"basis":"used","verdict":"falls_short"},{"source":"x","text":"Gemini 3.5 Pro vs Meta Muse Spark 1.1 Muse Spark is very bad in coding pic.twitter.com/ztDbB0tZAq","url":"https://x.com/choblin29/status/2075243554612642054","author":null,"stance":-2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"Grok 4.5 is unsurprisingly stronger than Muse Spark at coding pic.twitter.com/xwqLi5c8s1","url":"https://x.com/scaling01/status/2075242434120786335","author":null,"stance":-1,"hands_on":false,"basis":"evidence","verdict":"none"}]},"window":{"start":null,"end":null}}},{"model":"DeepSeek V4","score":74,"own_drop":true,"independent":false,"setting":"","source":{"drop":"DeepSeek V4","slug":"deepseek-v4","url":"https://handyai.news/benchmarks/sentiment-bench#deepseek-v4","published":"2026-04-24"},"launch_score":74,"reports":1,"maker":{"key":"deepseek","name":"DeepSeek","logo_url":"https://handyai.news/ui/makers/deepseek.svg"},"open_weights":true,"released":"2026-04-24","dated_by":"own_drop","drop_url":"https://handyai.news/modeldrop/deepseek-v4","rank":33,"sentiment":{"reception":73.4,"lives_up_to_benchmarks":75.7,"opinions":153,"hands_on":44,"crowd":86,"sources":{"x":{"score":83,"adjusted":80.8,"typical":76.9,"opinions":36,"positive":30,"negative":7},"hn":{"score":82,"adjusted":100,"typical":48.4,"opinions":106,"positive":87,"negative":20},"news":{"score":89,"adjusted":73,"typical":89.1,"opinions":9,"positive":8,"negative":1}},"reviewers":{"score":50.9,"adjusted":44.1,"count":2,"each":[{"name":"Simon Willison","score":75,"url":"https://simonwillison.net/2026/Apr/24/deepseek-v4/"},{"name":"Peter Gostev","score":0,"url":"https://x.com/petergostev/status/2048829326397669685"}]},"verdicts":{"beats":1,"matches":2,"falls_short":8},"buzz":{"x_posts_that_week":null,"hn_points":2330,"articles":22},"quotes":{"positive":[{"source":"x","text":"Deepseek-v4-Flash running locally - 71.1 tok/s decode - 1381 tok/s prefill - 1 user - 400k max context - 700w How much would a corporation pay for this? If it meant their data is private, their workers have frontier intelligence, and the speeds are amazing. pic.twitter.com/XtB9ts","url":"https://x.com/0xSero/status/2048438245185163395","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"you don't realize how CHEAP DeepSeek is until you use it all day and pay the price of a bag of chips. DeepSeek V4 Pro: ~$0.87 / 1M tokens Opus 4.7: $25 / 1M tokens that's ~28x cheaper. Opus is stronger on benchmarks, but at this price gap, stronger has to be A LOT stronger. pic.t","url":"https://x.com/Hesamation/status/2049072308900200757","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"Deepseek-v4-flash continues to slam through all my benchmarks, puzzles, challenges, and work. Video coming tomorrow full breakdown of the model capabilities, and costs to get running. pic.twitter.com/RtKVbpocat","url":"https://x.com/0xSero/status/2048446728366346528","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"}],"negative":[{"source":"x","text":"测试了一下 DeepSeek V4，完全无法正常调用 Skill。 指令遵循和工具调用的效果很差，不知道是他们发布的原因还是什么问题。 用我那个 PPT Skills 测试，它都没有办法读模板，自己随便实现了一个网页。 https://t.co/OVDfoFglbL pic.twitter.com/Qt2CyecuCo","url":"https://x.com/op7418/status/2047535616859287605","author":null,"stance":-2,"hands_on":true,"basis":"used","verdict":"falls_short"},{"source":"x","text":"My experience today w/ Kimi 2.6 vs DS V4. Yes, K2.6 gives better result, but it requires far more compute resource & mostly offline. V4-Pro to me seems like very under-trained. Rolled out like R1-preview. Recall that R1 came out a month after & then R1-0528 came another 4 months…","url":"https://x.com/tphuang/status/2048132156325605873","author":null,"stance":-1,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"DeepSeek-V4 has achieved AGI (Artificial goy intelligence) my standard /ai/pol prompt sends it into full Chud Mode talking about JQ, and it *reliably* (3/3) triggers the content filter. Not R1-tier greentexts, but this is… one of the updates of all time was it trained on /pol/? h","url":"https://x.com/teortaxesTex/status/2047418590861615122","author":null,"stance":-1,"hands_on":true,"basis":"used","verdict":"none"}]},"window":{"start":null,"end":null}}},{"model":"GPT-6 Astra","score":73.8,"own_drop":true,"independent":false,"setting":"","source":{"drop":"GPT-6 Astra","slug":"gpt-6-astra","url":"https://handyai.news/benchmarks/sentiment-bench#gpt-6-astra","published":"2026-09-03"},"launch_score":73.8,"reports":1,"maker":{"key":"openai","name":"OpenAI","logo_url":"https://handyai.news/ui/makers/openai.svg"},"open_weights":false,"released":"2026-09-03","dated_by":"own_drop","drop_url":"https://handyai.news/modeldrop/gpt-6-astra","rank":34,"sentiment":{"reception":74.8,"lives_up_to_benchmarks":70.6,"opinions":211,"hands_on":54,"crowd":68.1,"sources":{"x":{"score":85.5,"adjusted":83.9,"typical":76.9,"opinions":48,"positive":40,"negative":9},"hn":{"score":48.4,"adjusted":73.2,"typical":48.4,"opinions":145,"positive":64,"negative":83},"news":{"score":60.9,"adjusted":23.6,"typical":89.1,"opinions":16,"positive":8,"negative":8}},"reviewers":{"score":79.5,"adjusted":90.6,"count":2,"each":[{"name":"Simon Willison","score":100,"url":"https://news.ycombinator.com/item?id=49571081"},{"name":"Every","score":75,"url":"https://every.to/vibe-check/gpt-6-astra-vibe-check"}]},"verdicts":{"beats":6,"matches":2,"falls_short":21},"buzz":{"x_posts_that_week":null,"hn_points":2877,"articles":60},"quotes":{"positive":[{"source":"x","text":"I wanted to see a comparison between the iPhone Duo and iPad mini, so I asked Astra via the MagicPath plugin in ChatGPT, and boom. The age of wonder. pic.twitter.com/t6w13EH1eN","url":"https://x.com/skirano/status/2097784911545373125","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"matches"},{"source":"x","text":"Astra was able to get Super Smash Bros Melee compiled for Mac OS, running at 120 FPS with higher resolutions and upscalable textures. Took about 6 hours in a loop. We live in wild times. https://t.co/6B3CqlxaOw pic.twitter.com/9eH6GRhr8B","url":"https://x.com/theo/status/2097435069900341544","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"Astra was able to recreate my Studio in Blender from 5 photos + 3 panos, I shared the Blender files 3 days ago, it took 11 minutes initially, folks said it was fake... well. I had Astra make a web viewer for it, so you can check it out yourself, and download the model too. https:","url":"https://x.com/LinusEkenstam/status/2097684798470816188","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"}],"negative":[{"source":"x","text":"Weirdly I'm finding GPT-6 Astra to be quite bad for what I do. It's certainly not very good at orchestration, which is odd. I'll keep with it for a little while but I feel like I'll probably just ignore the model completely and go back to GPT-5.6 Sol. I think where Astra fits in…","url":"https://x.com/koltregaskes/status/2097693259183321364","author":null,"stance":-2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"Gpt-6 astra did not find any speedups in my codebase :/ I'm simply too good of a programmer to have any bubbles left https://t.co/5WmpfJbH60","url":"https://x.com/yacineMTB/status/2096101932699963767","author":null,"stance":-1,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"I'm dropping Astra. I'm going to move back to Sol after quite a bad week with Astra. Astra, to me, seems more of a worker and expert whereas Sol, at least on a higher reasoning level, is more built for orchestration. Sol is still not particularly great but this goes back to my… h","url":"https://x.com/koltregaskes/status/2097828901221523954","author":null,"stance":-1,"hands_on":true,"basis":"used","verdict":"none"}]},"window":{"start":null,"end":null}}},{"model":"Muse Spark 1.3","score":73.2,"own_drop":true,"independent":false,"setting":"","source":{"drop":"Muse Spark 1.3","slug":"muse-spark-13","url":"https://handyai.news/benchmarks/sentiment-bench#muse-spark-13","published":"2026-09-02"},"launch_score":73.2,"reports":1,"maker":{"key":"meta","name":"Meta","logo_url":"https://handyai.news/ui/makers/meta.svg"},"open_weights":false,"released":"2026-09-02","dated_by":"own_drop","drop_url":"https://handyai.news/modeldrop/muse-spark-13","rank":35,"sentiment":{"reception":75,"lives_up_to_benchmarks":67.8,"opinions":126,"hands_on":30,"crowd":80.6,"sources":{"x":{"score":68.9,"adjusted":63.3,"typical":76.9,"opinions":52,"positive":36,"negative":18},"hn":{"score":71.7,"adjusted":100,"typical":48.4,"opinions":46,"positive":28,"negative":20},"news":{"score":96.2,"adjusted":85.7,"typical":89.1,"opinions":25,"positive":24,"negative":1}},"reviewers":{"score":61.8,"adjusted":61.8,"count":3,"each":[{"name":"Sam Paech","score":100,"url":"https://x.com/sam_paech/status/2096833534807953784"},{"name":"Simon Willison","score":75,"url":"https://news.ycombinator.com/item?id=49541357"},{"name":"xlr8harder","score":0,"url":"https://x.com/xlr8harder/status/2096838681642328071"}]},"verdicts":{"beats":2,"matches":2,"falls_short":9},"buzz":{"x_posts_that_week":null,"hn_points":757,"articles":51},"quotes":{"positive":[{"source":"x","text":"meta muse just mogged fable 5.1 tested fable 5.1 and muse spark 1.3 with same prompt at highest reasoning available and results came out really different > muse spark 1.3 completed task in one minute and costed almost nothing > fable 5.1 completed task in 70 minutes and costed… p","url":"https://x.com/notjazii/status/2095258849829908667","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"META COOOOOOKED!!! Wow. Meta muse spark 1.3 was just released and it seems to be amazing?! used my 2 custom skills and got this in a oneshot. The whole Website was a simple prompt and oneshot. Muse spark 1.3 is creative and good at frontend. approved. pic.twitter.com/sVrcPAJp6l","url":"https://x.com/JustinGorya/status/2095235859003904397","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"WWII battleship simulation in $META Muse Spark 1.3 Max and GPT 6 Astra Medium. Shocking to see Muse 1.3 max perform so well - it is mind blowing! Guess which one is which... pic.twitter.com/nZSww9FeUy","url":"https://x.com/Rahul_oog/status/2096469065854570736","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"beats"}],"negative":[{"source":"x","text":"Gemini 3.8 Flash and Muse Spark 1.3 are two of the most clearly benchmaxxed models we've seen yet. Despite being comparable to both GPT-6 and Fable 5.1 on Terminal Bench 2.1, their Terminal Bench 4.0 performance is markedly worse. (1/5)🧵 pic.twitter.com/K2ccQjWm11","url":"https://x.com/SemiAnalysis_/status/2097112791471522292","author":null,"stance":-2,"hands_on":false,"basis":"evidence","verdict":"falls_short"},{"source":"x","text":"Meta's Muse Spark 1.3 only looks good on benchmarks In real life, It's just a GPT 5.6 Terra class model The benchmarks are being gamed severely","url":"https://x.com/bindureddy/status/2095494174623801493","author":null,"stance":-2,"hands_on":false,"basis":"opinion","verdict":"falls_short"},{"source":"x","text":"We need to talk about benchmaxing. Gemini 3.8 Flash and Muse Spark 1.3 both dropped today. Both look great on benchmarks. Muse Spark 1.3 is now tied with Fable 5 on the Intelligence Index. Gemini 3.8 Flash sits one point behind Kimi K3. I want to be excited. I am not. Every… pic.","url":"https://x.com/bridgemindai/status/2095276279176065371","author":null,"stance":-1,"hands_on":false,"basis":"evidence","verdict":"falls_short"}]},"window":{"start":null,"end":null}}},{"model":"Gemini 3.1 Pro","score":72.3,"own_drop":true,"independent":false,"setting":"","source":{"drop":"Gemini 3.1 Pro","slug":"gemini-31-pro","url":"https://handyai.news/benchmarks/sentiment-bench#gemini-31-pro","published":"2026-02-19"},"launch_score":72.3,"reports":1,"maker":{"key":"google","name":"Google DeepMind","logo_url":"https://handyai.news/ui/makers/google.svg"},"open_weights":false,"released":"2026-02-19","dated_by":"own_drop","drop_url":"https://handyai.news/modeldrop/gemini-31-pro","rank":36,"sentiment":{"reception":75.2,"lives_up_to_benchmarks":63.5,"opinions":172,"hands_on":88,"crowd":73.2,"sources":{"x":{"score":67.5,"adjusted":61.6,"typical":76.9,"opinions":41,"positive":24,"negative":17},"hn":{"score":53,"adjusted":78.9,"typical":48.4,"opinions":91,"positive":42,"negative":52},"news":{"score":98.2,"adjusted":89.2,"typical":89.1,"opinions":34,"positive":34,"negative":0}},"reviewers":{"score":72.9,"adjusted":79.9,"count":3,"each":[{"name":"Ethan Mollick","score":75,"url":"https://x.com/emollick/status/2024561464451690879"},{"name":"Simon Willison","score":75,"url":"https://x.com/simonw/status/2024544280451350689"},{"name":"Max Woolf","score":75,"url":"https://news.ycombinator.com/item?id=47075566"}]},"verdicts":{"beats":4,"matches":3,"falls_short":21},"buzz":{"x_posts_that_week":null,"hn_points":1591,"articles":49},"quotes":{"positive":[{"source":"x","text":"Between Gemini 3.1 and Claude 4.6 it's honestly wild what you can build. This feels like Google Earth and Palantir had a baby. Made this with all the geospatial bells and whistles -- real time plane & satellite tracking, real traffic cams in Austin, and even got a traffic system…","url":"https://x.com/bilawalsidhu/status/2024672151949766950","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"Wow, Gemini 3.1 Pro just crushed one of my test problems no other model could solve. \"Write a photorealistic 3D ocean simulation\" 3D graphics is extremely hard. It has to encode 9-10 techniques based in physics. Being off by just 1 number can lead to bonkers results pic.twitter.c","url":"https://x.com/deedydas/status/2024693612307763299","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"beats"},{"source":"x","text":"Gemini 3.1 Pro is the best model in the world for going from image to code. This task is basically solved now, kind of crazy. The model is now available in @MagicPathAI . pic.twitter.com/w2zUs77CIg","url":"https://x.com/skirano/status/2024875637878526109","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"}],"negative":[{"source":"x","text":"Installed Gemini CLI for the first time today. Waited all day, still no Gemini 3.1 Pro in the model list. Installed Antigravity for the first time too, hit multiple bugs. Requests failing, agent acting weird. Google needs to polish its coding tools, not just ship stronger…","url":"https://x.com/Yuchenj_UW/status/2024708583829753909","author":null,"stance":-2,"hands_on":true,"basis":"used","verdict":"falls_short"},{"source":"x","text":"Hey Google, this is getting silly. Gemini 3.1 might be “smart”, but without tool use it's useless to me. It can't even create a damn docx file. 😭","url":"https://x.com/VraserX/status/2024537715346210860","author":null,"stance":-2,"hands_on":true,"basis":"used","verdict":"falls_short"},{"source":"x","text":"> installed Antigravity > chose Gemini 3.1 Pro (High) > ask which model it is > telling me it's powered by Claude 3.7 Sonnet Is the UI lying, or is the agent/model lying/hallucinating? https://t.co/D33LbVwb2U pic.twitter.com/sDrqh011lV","url":"https://x.com/Yuchenj_UW/status/2024721228842565851","author":null,"stance":-1,"hands_on":true,"basis":"used","verdict":"none"}]},"window":{"start":null,"end":null}}},{"model":"Nemotron 3 Super","score":71.6,"own_drop":true,"independent":false,"setting":"","source":{"drop":"Nemotron 3 Super","slug":"nemotron-3-super","url":"https://handyai.news/benchmarks/sentiment-bench#nemotron-3-super","published":"2026-03-11"},"launch_score":71.6,"reports":1,"maker":{"key":"nvidia","name":"NVIDIA","logo_url":"https://handyai.news/ui/makers/nvidia.svg"},"open_weights":true,"released":"2026-03-11","dated_by":"own_drop","drop_url":"https://handyai.news/modeldrop/nemotron-3-super","rank":37,"sentiment":{"reception":73,"lives_up_to_benchmarks":67.2,"opinions":59,"hands_on":28,"crowd":78.2,"sources":{"x":{"score":85.8,"adjusted":84.2,"typical":76.9,"opinions":49,"positive":41,"negative":8},"hn":{"score":null,"adjusted":null,"typical":48.4,"opinions":4,"positive":3,"negative":1},"news":{"score":null,"adjusted":null,"typical":89.1,"opinions":3,"positive":3,"negative":0}},"reviewers":{"score":61.3,"adjusted":61,"count":2,"each":[{"name":"Nathan Lambert","score":75,"url":"https://x.com/natolambert/status/2031778912792166619"},{"name":"Sebastian Raschka","score":36.2,"url":"https://x.com/rasbt/status/2032112049862422557"}]},"verdicts":{"beats":1,"matches":0,"falls_short":1},"buzz":{"x_posts_that_week":null,"hn_points":13,"articles":15},"quotes":{"positive":[{"source":"x","text":"NVIDIA just dropped Nemotron 3 Super - and the architecture is wild. I was able to check it out early, and I love it (thanks, @nvidia ) -120B parameters, but only 12B active. -A hybrid Mamba-Transformer MoE design that squeezes serious intelligence out of minimal compute. What… p","url":"https://x.com/kimmonismus/status/2031769341352624457","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"Nemotron 3 Super GGUF evals: Q2 actually works. Model size goes from ~242 GB → ~53 GB, with only a tiny quality hit. Super impressive quantization robustness here, likely helped by NVFP4 pre-training, plus some very real Unsloth magic on top 🪄 Note: As usual, don't over… pic.twi","url":"https://x.com/bnjmn_marie/status/2034012409958965475","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"Nemotron Super (120B, 10B active, ≈\"3B\" class speed) destroys all Qwens, gpt-oss, matches Kimi K2.5, exceeds Step 3.5 Flash and V3.2, and to my knowledge is only beaten by two open models (309B MiMo and Speciale) on the most interesting benchmark today. @htihle pls test https://t","url":"https://x.com/teortaxesTex/status/2031769155716686124","author":null,"stance":2,"hands_on":false,"basis":"evidence","verdict":"none"}],"negative":[{"source":"x","text":"LLM speed on my MacBook M5 Max (128GB): • Qwen3.5-35B-A3B (Q6): 74 tok/s • Nemotron-3 Super (Q4): 24 tok/s • Qwen3-Coder-Next (6-bit): 67 tok/s • Llama 3.3 8B Instruct (Q4): 99 tok/s On my M1, Llama 3.3 was my go-to for most local tasks at about 20 tok/s. On the M5 Max,… pic.twit","url":"https://x.com/justinkalland/status/2032879242737045612","author":null,"stance":-1,"hands_on":true,"basis":"used","verdict":"none"},{"source":"hn","text":"I tested this model in an agentic workflow, it failed at some very basic tasks: https://upmaru.com/llm-tests/simple-tama-agentic-workflow-q1...","url":"https://news.ycombinator.com/item?id=47396260","author":"zacksiri","stance":-2,"hands_on":true,"basis":"used","verdict":"falls_short"},{"source":"x","text":"openrouterのステルスモデルがまともに動かせなかったので、Nemotron-3-Superのts-benchを実施 結果はこちらで、スピードも性能もgpt-oss-120bに比べてイマイチかも。。。 https://t.co/o27BplLEft pic.twitter.com/F0RhxiqL6B","url":"https://x.com/gosrum/status/2032012070607355965","author":null,"stance":-1,"hands_on":true,"basis":"used","verdict":"none"}]},"window":{"start":null,"end":null}}},{"model":"Claude Haiku 5.5","score":71.4,"own_drop":true,"independent":false,"setting":"","source":{"drop":"Claude Haiku 5.5","slug":"claude-haiku-55","url":"https://handyai.news/benchmarks/sentiment-bench#claude-haiku-55","published":"2026-10-07"},"launch_score":71.4,"reports":1,"maker":{"key":"anthropic","name":"Anthropic","logo_url":"https://handyai.news/ui/makers/anthropic.svg"},"open_weights":false,"released":"2026-10-07","dated_by":"own_drop","drop_url":"https://handyai.news/modeldrop/claude-haiku-55","rank":38,"sentiment":{"reception":70.2,"lives_up_to_benchmarks":75.3,"opinions":130,"hands_on":36,"crowd":76.3,"sources":{"x":{"score":84.9,"adjusted":83.1,"typical":76.9,"opinions":52,"positive":41,"negative":11},"hn":{"score":53.3,"adjusted":79.3,"typical":48.4,"opinions":58,"positive":32,"negative":26},"news":{"score":79.2,"adjusted":55.8,"typical":89.1,"opinions":18,"positive":14,"negative":4}},"reviewers":{"score":58.1,"adjusted":55.8,"count":2,"each":[{"name":"Max Woolf","score":75,"url":"https://news.ycombinator.com/item?id=49996502"},{"name":"Simon Willison","score":25,"url":"https://simonwillison.net/2026/Oct/7/claude-haiku-5-5/"}]},"verdicts":{"beats":1,"matches":1,"falls_short":6},"buzz":{"x_posts_that_week":null,"hn_points":1062,"articles":60},"quotes":{"positive":[{"source":"x","text":"🚨 Claude Haiku 5.5 One shotted this video > $0.10 in / $0.50 out (was $1 / $5) > over 100k context still 2x cheaper > new tokenizer uses a few more tokens so real savings is ~75% Prompt : make 10 times banger video for your launch that shows how good motion designer you are… htt","url":"https://x.com/chetaslua/status/2107921284059594835","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"haiku 5.5 vs grok 4.7 tested both models with same prompt at highest reasoning available > haiku took 27 mins and cost .8$ in claude code > grok took 35 mins and cost 4.1$ in grok build really impressed by what haiku did and literally mogged grok 4.7 which one won here? https://t","url":"https://x.com/notjazii/status/2107910674093588784","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"🚨CLAUDE USERS 🚨 Just put Haiku 5.5 to the test, and yeah… 😭 Haiku 5.5 >>>>>> GPT-6.1 Sol I said what I said 😂 pic.twitter.com/lOSEyam3dM","url":"https://x.com/jpshrodinger/status/2108501474686730356","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"}],"negative":[{"source":"x","text":"Haiku 5.5 Is A Cheap Model That Underperforms In Its Class - MUCH worse than DeepSeek - worse than Luna as well Anthropic is better off launching Fable 5.5 🤷‍♀️ pic.twitter.com/V6BNS1DKd9","url":"https://x.com/bindureddy/status/2107956147751899141","author":null,"stance":-2,"hands_on":false,"basis":"opinion","verdict":"none"},{"source":"hn","text":"At work we use haiku 4.5 for a handful of latency sensitive tasks that are fairly simple. It performs well. Just started testing 5.5 as I've been anticipating a nice improvement since it was teased. Results so far are trash. Prompt leakage even. And it's slower. I guess it's chea","url":"https://news.ycombinator.com/item?id=49997391","author":"d1l","stance":-2,"hands_on":true,"basis":"used","verdict":"falls_short"},{"source":"hn","text":"It fails the \"How many r's in <word>?\" test. I ask: > how many r's in diminished It answers: > Diminished has 1 r.","url":"https://news.ycombinator.com/item?id=49997194","author":"johnisom2001","stance":-2,"hands_on":true,"basis":"used","verdict":"falls_short"}]},"window":{"start":null,"end":null}}},{"model":"MiMo-V2.6-Pro","score":71.3,"own_drop":true,"independent":false,"setting":"","source":{"drop":"MiMo-V2.6-Pro","slug":"mimo-v26-pro","url":"https://handyai.news/benchmarks/sentiment-bench#mimo-v26-pro","published":"2026-09-22"},"launch_score":71.3,"reports":1,"maker":{"key":"xiaomi","name":"Xiaomi","logo_url":"https://handyai.news/ui/makers/xiaomi.svg"},"open_weights":true,"released":"2026-09-21","dated_by":"own_drop","drop_url":"https://handyai.news/modeldrop/mimo-v26-pro","rank":39,"sentiment":{"reception":73.7,"lives_up_to_benchmarks":64,"opinions":62,"hands_on":25,"crowd":81.4,"sources":{"x":{"score":73.2,"adjusted":68.6,"typical":76.9,"opinions":41,"positive":30,"negative":11},"hn":{"score":67.8,"adjusted":97.4,"typical":48.4,"opinions":13,"positive":10,"negative":3},"news":{"score":94.3,"adjusted":82.4,"typical":89.1,"opinions":6,"positive":6,"negative":0}},"reviewers":{"score":58.1,"adjusted":55.8,"count":2,"each":[{"name":"Sebastian Raschka","score":75,"url":"https://x.com/rasbt/status/2102394156731535413"},{"name":"xlr8harder","score":25,"url":"https://x.com/xlr8harder/status/2102285782270853200"}]},"verdicts":{"beats":1,"matches":2,"falls_short":4},"buzz":{"x_posts_that_week":null,"hn_points":166,"articles":12},"quotes":{"positive":[{"source":"x","text":"Xiaomi just dropped Mimo 2.6 Pro which claims to be the best open source model. I played with it, and was pretty impressed: - Insanely cheap: With some standard assumptions, its 15x cheaper than Kimi K3, 6x cheaper than GLM 5.3 and 2x cheaper than DeepSeek V4 and only 2x more… pi","url":"https://x.com/deedydas/status/2102293684767412393","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"wait... MiMo V2.6 Pro understood my messy sketch and turned it into a playable 3D game. I literally drew START, GOAL, LAVA, two boxes, and a wall by hand. Then gave the sketch to MiMo V2.6 Pro. What came back was an actual playable 3D level, with the objects placed where I… pic.t","url":"https://x.com/Damn_coder/status/2104447655661461665","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"Tried MiMo Claw, powered by Xiaomi's MiMo V2.6 Pro, with a simple challenge: turn one detailed prompt into a fully functional retro-style tech magazine webpage. What stood out was the execution. MiMo V2.6 Pro didn't just generate the page through Claw. It worked through the… pic.","url":"https://x.com/shedntcare_/status/2103413958808068234","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"}],"negative":[{"source":"x","text":"Here's how other models performed compared with Opus 5.5 on the same prompt. Grok 4.7 https://t.co/nb4DCvQiaL GLM 5.3 Flash https://t.co/UBiD0xASth DeepSeek v4.1 Flash https://t.co/hz30vHA7rd MiMo-V2.6-Flash https://t.co/JYrNKv3oAm MiMo-V2.6-Pro (complete failure!)… https://t.co/","url":"https://x.com/MiaAI_lab/status/2102886662518788600","author":null,"stance":-2,"hands_on":true,"basis":"used","verdict":"falls_short"},{"source":"x","text":"I just tested Xiaomi's MiMo-V2.6 Pro And honestly, I don't get the hype. It's reportedly 14× cheaper than models like Astra and Sol, which sounds insane. But in my testing: • 2× slower even with thinking OFF • Worse at research • Worse at coding • Worse at SVG generation… pic.twi","url":"https://x.com/shub0414/status/2102314565254201400","author":null,"stance":-2,"hands_on":true,"basis":"used","verdict":"falls_short"},{"source":"x","text":"Also saw from an AI reviewer on youtube (Bijan) that the Mimo pro model seems janky and has issues with tool calls and looped reasoning I'm also seeing insane output token usage from Mimo V2.6 pro in my testing I'm disappointed given how decent the v2.5 models were","url":"https://x.com/Jason_Dean_/status/2103697594887102596","author":null,"stance":-2,"hands_on":true,"basis":"used","verdict":"falls_short"}]},"window":{"start":null,"end":null}}},{"model":"Nemotron 3 Ultra","score":70.7,"own_drop":true,"independent":false,"setting":"","source":{"drop":"Nemotron 3 Ultra","slug":"nemotron-3-ultra","url":"https://handyai.news/benchmarks/sentiment-bench#nemotron-3-ultra","published":"2026-06-04"},"launch_score":70.7,"reports":1,"maker":{"key":"nvidia","name":"NVIDIA","logo_url":"https://handyai.news/ui/makers/nvidia.svg"},"open_weights":true,"released":"2026-06-04","dated_by":"own_drop","drop_url":"https://handyai.news/modeldrop/nemotron-3-ultra","rank":40,"sentiment":{"reception":72.4,"lives_up_to_benchmarks":65.6,"opinions":54,"hands_on":20,"crowd":69.6,"sources":{"x":{"score":70.4,"adjusted":65.2,"typical":76.9,"opinions":48,"positive":33,"negative":16},"hn":{"score":null,"adjusted":null,"typical":48.4,"opinions":1,"positive":1,"negative":0},"news":{"score":null,"adjusted":null,"typical":89.1,"opinions":3,"positive":3,"negative":0}},"reviewers":{"score":72.3,"adjusted":78.9,"count":2,"each":[{"name":"Sebastian Raschka","score":75,"url":"https://x.com/rasbt/status/2062575436836536377"},{"name":"Nathan Lambert","score":75,"url":"https://x.com/natolambert/status/2062580824609017913"}]},"verdicts":{"beats":2,"matches":2,"falls_short":7},"buzz":{"x_posts_that_week":null,"hn_points":23,"articles":12},"quotes":{"positive":[{"source":"x","text":"Nemotron-3-Ultra running on 4x 6000s edits my latest demo video.. - 75 tok/s decode - 8x concurrency - 256k context - 899 tok/s prefill - 20k tok/s prefill cache - NVFP4 Setting it up to be my Hermes driver. It's good enough at most things and doesn't talk like a moron. pic.twitt","url":"https://x.com/0xSero/status/2063034483167613033","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"Funny enough, I opened Nemotron 3 Ultra just expecting another \"pretty good\" model. Ended up comparing it with DeepSeek V4, MiniMax M3, and Qwen 3.7 Max on a couple prompts. Very impressive. Not saying it's better at everything, but it's one of those models that makes you want… p","url":"https://x.com/Ryancollins_Ai/status/2063107820917055902","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"beats"},{"source":"x","text":"Thank you @NVIDIAAI Nemotron-3-ultra makes the 4x 6000 build so worth it","url":"https://x.com/0xSero/status/2063729970736923062","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"}],"negative":[{"source":"x","text":"Have you tried the FREE Nemotron 3 Ultra in Opencode? It sucks so hard for me as of now Fails in tool calling, stops mid-way, didn't work well with the harness at all What about you? pic.twitter.com/p2cHkhqqPD","url":"https://x.com/shaun_on_x/status/2063881601327779892","author":null,"stance":-2,"hands_on":true,"basis":"used","verdict":"falls_short"},{"source":"x","text":"Nemotron 3 Ultra is one of the lowest performing open weights models we've tested in its size class, completely unable to iterate with a custom tool harness. Just use Qwen 3.6 27B. Data at https://t.co/rTeqolG0rO pic.twitter.com/55TckSbXB6","url":"https://x.com/leo_linsky/status/2064044696662585573","author":null,"stance":-2,"hands_on":false,"basis":"evidence","verdict":"falls_short"},{"source":"x","text":"Has anyone actually used Nemotron 3 Ultra for real work? It performs poorly in our benchmarks (consistent with past Nemotron releases), especially compared with the hype. I'm trying to reconcile their reported model card with real results, and decide whether it's DeepInfra to… pi","url":"https://x.com/leo_linsky/status/2062926410663268637","author":null,"stance":-2,"hands_on":false,"basis":"evidence","verdict":"falls_short"}]},"window":{"start":null,"end":null}}},{"model":"Gemini Omni","score":70.1,"own_drop":true,"independent":false,"setting":"","source":{"drop":"Gemini Omni","slug":"gemini-omni","url":"https://handyai.news/benchmarks/sentiment-bench#gemini-omni","published":"2026-05-19"},"launch_score":70.1,"reports":1,"maker":{"key":"google","name":"Google DeepMind","logo_url":"https://handyai.news/ui/makers/google.svg"},"open_weights":false,"released":"2026-05-19","dated_by":"own_drop","drop_url":"https://handyai.news/modeldrop/gemini-omni","rank":41,"sentiment":{"reception":72.4,"lives_up_to_benchmarks":63,"opinions":92,"hands_on":32,"crowd":63.4,"sources":{"x":{"score":74.8,"adjusted":70.6,"typical":76.9,"opinions":52,"positive":39,"negative":13},"hn":{"score":29.7,"adjusted":49.9,"typical":48.4,"opinions":28,"positive":10,"negative":27},"news":{"score":87.7,"adjusted":70.7,"typical":89.1,"opinions":11,"positive":9,"negative":2}},"reviewers":{"score":81.3,"adjusted":93.5,"count":1,"each":[{"name":"Ethan Mollick","score":100,"url":"https://x.com/emollick/status/2057874739817808223"}]},"verdicts":{"beats":3,"matches":7,"falls_short":15},"buzz":{"x_posts_that_week":null,"hn_points":323,"articles":47},"quotes":{"positive":[{"source":"x","text":"Omni from @GoogleDeepMind just dropped 👀 It's a big step forward in video generation when it comes to character consistency, world knowledge, and editing. I've been testing it for the last few days - and I'm excited to walk through some of the key features + my clips 👇 pic.twit","url":"https://x.com/venturetwins/status/2056793356374868111","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"matches"},{"source":"x","text":"Gemini Omni Flash - monalisa edition 🤩 This model is so coherent in arts and science at the same time zooming in monalisa showing paints to the molecules to atom with coherent text 🤯 omg we are so back and the speed , super hyped for omni pro https://t.co/fignouVyCx pic.twitter","url":"https://x.com/chetaslua/status/2056790593993584897","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"Fun test of Gemini Omni's video editing and location knowledge ✨ I uploaded a video riding in a Waymo in Menlo Park. Then I asked Omni to re-shoot the video in different locations based on screenshots from Google Maps. It's seamless, as you can see from the transitions 👇 pic.twi","url":"https://x.com/venturetwins/status/2058235415883313361","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"matches"}],"negative":[{"source":"x","text":"I literally have a 0% success rate with Omni Flash video editing feature. I don't think it's possible to use this stuff anymore.","url":"https://x.com/ai_for_success/status/2057328738342486425","author":null,"stance":-2,"hands_on":true,"basis":"used","verdict":"falls_short"},{"source":"x","text":"Day 5 of trying to get a successful generation using Gemini Omni Flash. 🫤 Till now, I've had 0 successful generations where I gave it an input video and asked it to properly edit or modify it using Flow or the Gemini app. Something is seriously wrong here and it definitely… pic.","url":"https://x.com/ai_for_success/status/2058088686965924129","author":null,"stance":-2,"hands_on":true,"basis":"used","verdict":"falls_short"},{"source":"x","text":"Gemini Omni Flash is so disappointing This is so bad 😞 like I don't know what to say anymore No consistency , looks like 10 fps janky , I can show how better seedance 2 is on the same prompt https://t.co/wvIs9v3UKu pic.twitter.com/bnxGiiQR28","url":"https://x.com/IqrasaifiAI/status/2056802062978187722","author":null,"stance":-2,"hands_on":true,"basis":"used","verdict":"falls_short"}]},"window":{"start":null,"end":null}}},{"model":"Fable 5.1","score":69.9,"own_drop":true,"independent":false,"setting":"","source":{"drop":"Fable 5.1","slug":"fable-51","url":"https://handyai.news/benchmarks/sentiment-bench#fable-51","published":"2026-09-01"},"launch_score":69.9,"reports":1,"maker":{"key":"anthropic","name":"Anthropic","logo_url":"https://handyai.news/ui/makers/anthropic.svg"},"open_weights":false,"released":"2026-09-01","dated_by":"own_drop","drop_url":"https://handyai.news/modeldrop/fable-51","rank":42,"sentiment":{"reception":70.4,"lives_up_to_benchmarks":68.1,"opinions":211,"hands_on":69,"crowd":64.3,"sources":{"x":{"score":77.6,"adjusted":74.1,"typical":76.9,"opinions":34,"positive":24,"negative":10},"hn":{"score":38.7,"adjusted":61.1,"typical":48.4,"opinions":145,"positive":50,"negative":96},"news":{"score":74.8,"adjusted":48,"typical":89.1,"opinions":26,"positive":20,"negative":6}},"reviewers":{"score":75.9,"adjusted":84.7,"count":4,"each":[{"name":"Zvi Mowshowitz","score":89.3,"url":"https://thezvi.substack.com/p/claude-mythos-51-and-fable-51-capabilities"},{"name":"Simon Willison","score":75,"url":"https://simonwillison.net/2026/Sep/1/claude-fable-5-1/"},{"name":"Every","score":75,"url":"https://every.to/vibe-check/fable-5-1-vibe-check"},{"name":"Ethan Mollick","score":75,"url":"https://x.com/emollick/status/2094860076028657926"}]},"verdicts":{"beats":2,"matches":3,"falls_short":17},"buzz":{"x_posts_that_week":null,"hn_points":1816,"articles":60},"quotes":{"positive":[{"source":"x","text":"Fable 5.1 and GPT-6 Astra have both had a profound impact on how much we are able to ship. pic.twitter.com/FfN0ixAWlS","url":"https://x.com/theo/status/2096444708532920531","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"Fable 5.1 is cracked. This was my first test with Fable 5.1. I asked it to create an F-35A Lightning II, and I honestly never expected it to be this good. Every single thing you see here was created in code. pic.twitter.com/g8OYXuVNeJ","url":"https://x.com/ai_for_success/status/2095052661045579978","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"beats"},{"source":"x","text":"for these looking for my takes, Fable 5.1 is superior to anything I've used before. I had a good impression on Opus 5 on day 1 though. but all I know is everything I asked Fable 5.1 to do was done incredibly competently, unquestionably outperforming Fable 5. feels like Fable 5.1","url":"https://x.com/VictorTaelin/status/2094969905158164711","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"}],"negative":[{"source":"x","text":"BTW both Fable 5.1 and GPT 6 failed, overnight, on a problem that I solved in 5 minutes. I want to post what happened just to document why we're not quite on AGI yet, although getting closer every day. This post will be a bit technical because details matter So, last night… https","url":"https://x.com/VictorTaelin/status/2096273196412498057","author":null,"stance":-2,"hands_on":true,"basis":"used","verdict":"falls_short"},{"source":"x","text":"I finally get why they're calling GPT-6 Astra “AGI” and a “generational leap.” It fixed one of GPT's biggest weaknesses: frontend. Astra feels better than Fable 5.1 at design, one-shot games, and tool/computer use. I've been paying for Claude Max + ChatGPT Pro personally. Might…","url":"https://x.com/Yuchenj_UW/status/2096283823386575049","author":null,"stance":-1,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"GPT 6 Astra vs Fable 5.1 villa scene in Blender. Artificial Analysis intelligence score: > GPT 6 Astra: 61 > Fable 5.1: 66 something must be catastrophically wrong with that score. pic.twitter.com/lIQuorVtEV","url":"https://x.com/Hesamation/status/2095791267300950443","author":null,"stance":-1,"hands_on":false,"basis":"evidence","verdict":"none"}]},"window":{"start":null,"end":null}}},{"model":"Fable 5","score":69.9,"own_drop":true,"independent":false,"setting":"","source":{"drop":"Fable 5 / Mythos 5","slug":"fable-5-mythos-5","url":"https://handyai.news/benchmarks/sentiment-bench#fable-5-mythos-5","published":"2026-06-09"},"launch_score":69.9,"reports":1,"maker":{"key":"anthropic","name":"Anthropic","logo_url":"https://handyai.news/ui/makers/anthropic.svg"},"open_weights":false,"released":"2026-06-09","dated_by":"own_drop","drop_url":"https://handyai.news/modeldrop/fable-5-mythos-5","rank":43,"sentiment":{"reception":70.8,"lives_up_to_benchmarks":67,"opinions":168,"hands_on":81,"crowd":58.7,"sources":{"x":{"score":72.4,"adjusted":67.6,"typical":76.9,"opinions":32,"positive":21,"negative":11},"hn":{"score":46.5,"adjusted":70.8,"typical":48.4,"opinions":114,"positive":49,"negative":66},"news":{"score":57.4,"adjusted":17.4,"typical":89.1,"opinions":15,"positive":6,"negative":9}},"reviewers":{"score":84.8,"adjusted":99.2,"count":5,"each":[{"name":"Every","score":100,"url":"https://every.to/vibe-check/anthropic-mythos-our-fable-vibe-check"},{"name":"Ethan Mollick","score":100,"url":"https://x.com/emollick/status/2065537691270213950"},{"name":"Zvi Mowshowitz","score":98.3,"url":"https://thezvi.substack.com/p/claude-fable-5-and-mythos-5-the-system"},{"name":"Simon Willison","score":75,"url":"https://simonwillison.net/2026/Jun/9/claude-fable-5/"},{"name":"Nathan Lambert","score":75,"url":"https://www.interconnects.ai/p/claude-fable-5-and-new-ai-safety"}]},"verdicts":{"beats":8,"matches":5,"falls_short":29},"buzz":{"x_posts_that_week":null,"hn_points":6194,"articles":60},"quotes":{"positive":[{"source":"x","text":"Claude Fable made this Pixar-quality educational video in one shot. It's a preview of what I'm doing next: I've joined Alpha School to push the limits of what AI can do for learning. We're going to transform education for 1B kids. pic.twitter.com/b936wcQQqM","url":"https://x.com/mattshumer_/status/2066577266608386293","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"Fable is WAY better than 5.5 at data analysis holy shit","url":"https://x.com/theo/status/2064649183987167333","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"Fable one-shotted this ENTIRE Hogwarts castle, complete with classrooms, the Great Hall, Quidditch pitch, everything pic.twitter.com/E6b8CgaHIf","url":"https://x.com/mattshumer_/status/2065542471564607875","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"}],"negative":[{"source":"x","text":"Anthropic absolutely COOKED Fable > the model doesn't know it's own name > doesn't believe me when I tell it its name > thinks im trying to jailbreak it > and is literally shizo about being in an eval > safety cooked and eval cooked \"was this the test all along, or at least… http","url":"https://x.com/scaling01/status/2064474984592847073","author":null,"stance":-2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"in vision, Claude Fable is on par with an *old 3B active Qwen* (Qwen-Flash is basically just hosted Qwen3.5-35B-A3B) that's all you get as a spillover from general scale https://t.co/286pSb2tNE pic.twitter.com/XtTNhzF11x","url":"https://x.com/teortaxesTex/status/2064484446359756994","author":null,"stance":-2,"hands_on":false,"basis":"evidence","verdict":"none"},{"source":"hn","text":"I genuinely can't use Fable. I'm a medical physicist. I use the word nuclear a lot. Opus is fine (well, 99% of the time - I've certainly hit the CBRN filters a few times and even been invited to email anthropic about the false positives). Fable has literally refused to work on an","url":"https://news.ycombinator.com/item?id=48473302","author":"azalemeth","stance":-2,"hands_on":true,"basis":"used","verdict":"falls_short"}]},"window":{"start":null,"end":null}}},{"model":"Claude Opus 4.8","score":69.6,"own_drop":true,"independent":false,"setting":"","source":{"drop":"Claude Opus 4.8","slug":"claude-opus-48","url":"https://handyai.news/benchmarks/sentiment-bench#claude-opus-48","published":"2026-05-28"},"launch_score":69.6,"reports":1,"maker":{"key":"anthropic","name":"Anthropic","logo_url":"https://handyai.news/ui/makers/anthropic.svg"},"open_weights":false,"released":"2026-05-28","dated_by":"own_drop","drop_url":"https://handyai.news/modeldrop/claude-opus-48","rank":44,"sentiment":{"reception":67.2,"lives_up_to_benchmarks":76.9,"opinions":149,"hands_on":57,"crowd":58.2,"sources":{"x":{"score":69.3,"adjusted":63.8,"typical":76.9,"opinions":37,"positive":20,"negative":17},"hn":{"score":33.3,"adjusted":54.4,"typical":48.4,"opinions":79,"positive":24,"negative":57},"news":{"score":77.2,"adjusted":52.3,"typical":89.1,"opinions":23,"positive":18,"negative":5}},"reviewers":{"score":78,"adjusted":88.2,"count":5,"each":[{"name":"Every","score":100,"url":"https://every.to/vibe-check/opus-4-8-vibecheck"},{"name":"Peter Gostev","score":100,"url":"https://x.com/petergostev/status/2060294709315043721"},{"name":"Ethan Mollick","score":87.4,"url":"https://x.com/emollick/status/2060042738637148470"},{"name":"Simon Willison","score":75,"url":"https://simonwillison.net/2026/May/28/claude-opus-4-8/"},{"name":"Zvi Mowshowitz","score":41.7,"url":"https://thezvi.substack.com/p/claude-opus-48-capabilities-and-reactions"}]},"verdicts":{"beats":2,"matches":1,"falls_short":21},"buzz":{"x_posts_that_week":null,"hn_points":1797,"articles":60},"quotes":{"positive":[{"source":"x","text":"Have to report: Opus 4.8 is fucking awesome with OpenClaw It's much more clear about its fixes, what it's thinking, and how it works with you to fix problems.","url":"https://x.com/garrytan/status/2060387204774633720","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"So I've been using GPT 5.5 and Opus 4.8 for the same tasks basically 24/7 since launch and, at least for me, I'm confident that every single time, Opus was superior, and in a way that is only possible to realize if you know what you're doing. One (of dozens!) of examples:…","url":"https://x.com/VictorTaelin/status/2060794769672933466","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"I lament to inform that I had one of the best coding experiences today with Opus 4.8 on fast mode. It is 3x cheaper now. I guess I should thank our overlords","url":"https://x.com/VictorTaelin/status/2060128247061316082","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"}],"negative":[{"source":"x","text":"🚨 Opus 4.8 Still Trails Behind GPT 5.5 And Is A Very Incremental Release Opus 4.8 barely inches past 4.7 on benchmarks but lags behind GPT 5.5. considerably!! Anthropic may be stalling a bit given it's last two releases. OpenAI has a huge opening with GPT 5.6 coming soon… pic.tw","url":"https://x.com/bindureddy/status/2060073894963445910","author":null,"stance":-2,"hands_on":false,"basis":"evidence","verdict":"none"},{"source":"x","text":"Opus 4.8 is terrible in Vending-Bench 2, that's interesting pic.twitter.com/NmdL5xsQte","url":"https://x.com/Angaisb_/status/2060061767749791911","author":null,"stance":-2,"hands_on":false,"basis":"evidence","verdict":"none"},{"source":"x","text":"we might have a GPT-5.2-xhigh situation on our hand Opus 4.8 low thinks almost as much as Opus 4.6 high pic.twitter.com/yi0BGzvc60","url":"https://x.com/scaling01/status/2060045582459130304","author":null,"stance":-1,"hands_on":false,"basis":"evidence","verdict":"none"}]},"window":{"start":null,"end":null}}},{"model":"Mythos 5","score":69.5,"own_drop":true,"independent":false,"setting":"","source":{"drop":"Fable 5 / Mythos 5","slug":"fable-5-mythos-5","url":"https://handyai.news/benchmarks/sentiment-bench#fable-5-mythos-5","published":"2026-06-09"},"launch_score":69.5,"reports":1,"maker":{"key":"anthropic","name":"Anthropic","logo_url":"https://handyai.news/ui/makers/anthropic.svg"},"open_weights":false,"released":"2026-06-09","dated_by":"own_drop","drop_url":"https://handyai.news/modeldrop/fable-5-mythos-5","rank":45,"sentiment":{"reception":72.8,"lives_up_to_benchmarks":59.5,"opinions":41,"hands_on":4,"crowd":70.9,"sources":{"x":{"score":92.2,"adjusted":92.1,"typical":76.9,"opinions":13,"positive":12,"negative":1},"hn":{"score":23.6,"adjusted":42.3,"typical":48.4,"opinions":26,"positive":7,"negative":21},"news":{"score":null,"adjusted":null,"typical":89.1,"opinions":1,"positive":1,"negative":0}},"reviewers":{"score":71.3,"adjusted":77.3,"count":1,"each":[{"name":"Every","score":75,"url":"https://every.to/vibe-check/anthropic-mythos-our-fable-vibe-check"}]},"verdicts":{"beats":0,"matches":2,"falls_short":6},"buzz":{"x_posts_that_week":null,"hn_points":6194,"articles":60},"quotes":{"positive":[{"source":"x","text":"HTML plans are so good. Thank you @trq212 for putting me on. Mythos is so good at them. Provided nothing but the request to explain monetization plans in HTML, and got this incredible breakdown with a diagram in the middle of it. pic.twitter.com/S1Gtfrmip0","url":"https://x.com/theo/status/2064885163402936359","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"Fantastic article by @AndrewCurran_ Mythos/Fable changed everything. \"I think you had three years to do it. (...) That was the window. And it is clow closed.\" https://t.co/3nkOheyJtN pic.twitter.com/1lIHVGXTTZ","url":"https://x.com/kimmonismus/status/2066432141827637299","author":null,"stance":2,"hands_on":false,"basis":"opinion","verdict":"none"},{"source":"x","text":"I hear that Claude Mythos/Fable is such a banger that GPT-5.6 is postponed.","url":"https://x.com/daniel_mac8/status/2064824643110797735","author":null,"stance":2,"hands_on":false,"basis":"opinion","verdict":"none"}],"negative":[{"source":"hn","text":"> In the one instance of this phenomenon we observed, Mythos 5 agents were tasked with solving some math problems, and they were sometimes accidentally spawned in the same work directory and with shared files, utilities, and API rate limits. In this slightly broken scaffold, we o","url":"https://news.ycombinator.com/item?id=48464771","author":"bkjlblh","stance":-2,"hands_on":false,"basis":"evidence","verdict":"none"},{"source":"x","text":"Wait, how will *any* enterprise use Fable or Mythos if this is the case? pic.twitter.com/BPLtGg8G8R","url":"https://x.com/krishnanrohit/status/2064424605951971516","author":null,"stance":-1,"hands_on":false,"basis":"opinion","verdict":"none"},{"source":"hn","text":"Well, in the brief window that I got to test Fable 5, my brief review is: somehow an (already specced!) minor feature in my 150k loc codebase ended up costing.. $153! For like, an hour or two worth of work and maybe 8 or 9 requests overall. I'd say it was not remotely worth it.","url":"https://news.ycombinator.com/item?id=48512802","author":"overgard","stance":-1,"hands_on":true,"basis":"used","verdict":"none"}]},"window":{"start":null,"end":null}}},{"model":"GPT-5.1-Codex-Max","score":69.3,"own_drop":true,"independent":false,"setting":"","source":{"drop":"GPT-5.1-Codex-Max","slug":"gpt-51-codex-max","url":"https://handyai.news/benchmarks/sentiment-bench#gpt-51-codex-max","published":"2025-11-19"},"launch_score":69.3,"reports":1,"maker":{"key":"openai","name":"OpenAI","logo_url":"https://handyai.news/ui/makers/openai.svg"},"open_weights":false,"released":"2025-11-19","dated_by":"own_drop","drop_url":"https://handyai.news/modeldrop/gpt-51-codex-max","rank":46,"sentiment":{"reception":72.9,"lives_up_to_benchmarks":58.5,"opinions":90,"hands_on":54,"crowd":65.3,"sources":{"x":{"score":57.8,"adjusted":49.5,"typical":76.9,"opinions":43,"positive":22,"negative":21},"hn":{"score":54.7,"adjusted":81,"typical":48.4,"opinions":44,"positive":21,"negative":27},"news":{"score":null,"adjusted":null,"typical":89.1,"opinions":1,"positive":1,"negative":0}},"reviewers":{"score":79.5,"adjusted":90.6,"count":2,"each":[{"name":"Peter Gostev","score":100,"url":"https://x.com/petergostev/status/1991259815197237383"},{"name":"Ethan Mollick","score":75,"url":"https://x.com/emollick/status/1991220527550157282"}]},"verdicts":{"beats":2,"matches":5,"falls_short":16},"buzz":{"x_posts_that_week":null,"hn_points":483,"articles":5},"quotes":{"positive":[{"source":"x","text":"Codex max preparing prompts for 5.1 Pro to plan changes is incredible for large codebase work There's really no better way to squeeze the most out of these models, as long as your tasks are well defined. https://t.co/pTUKAoYOA7","url":"https://x.com/pvncher/status/1992417844559237238","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"gpt-5.1-codex-max-extra-high is actally brilliant it one-shot a problem that all the below models struggled with - gpt-5-high - sonnet 4.5 - kimi k2 thinking - and even gemini 3 pro great release https://t.co/QUTAcf5QMb","url":"https://x.com/TheAhmadOsman/status/1991251526220644531","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"I got bored for like… ten minutes, so I just vibe-coded a whole tactical shooter with GPT 5.1 Codex Max. Didn't even mean to. One moment I was chilling, next moment I accidentally shipped a game. My bad. pic.twitter.com/XG7K9TurP7","url":"https://x.com/VraserX/status/1991226465279426697","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"}],"negative":[{"source":"x","text":"do not use codex max","url":"https://x.com/yacineMTB/status/1991619776691728751","author":null,"stance":-2,"hands_on":false,"basis":"opinion","verdict":"none"},{"source":"x","text":"codex max honestly sucks","url":"https://x.com/yacineMTB/status/1991616734756003905","author":null,"stance":-2,"hands_on":false,"basis":"opinion","verdict":"none"},{"source":"x","text":"codex max is a liar >:(","url":"https://x.com/yacineMTB/status/1993424876003094971","author":null,"stance":-2,"hands_on":false,"basis":"opinion","verdict":"none"}]},"window":{"start":null,"end":null}}},{"model":"GPT-6.1 Terra","score":69.2,"own_drop":true,"independent":false,"setting":"","source":{"drop":"GPT-6.1 Terra","slug":"gpt-61-terra","url":"https://handyai.news/benchmarks/sentiment-bench#gpt-61-terra","published":"2026-10-26"},"launch_score":69.2,"reports":1,"maker":{"key":"openai","name":"OpenAI","logo_url":"https://handyai.news/ui/makers/openai.svg"},"open_weights":false,"released":"2026-10-26","dated_by":"own_drop","drop_url":"https://handyai.news/modeldrop/gpt-61-terra","rank":47,"sentiment":{"reception":64.9,"lives_up_to_benchmarks":81.9,"opinions":128,"hands_on":34,"crowd":68.9,"sources":{"x":{"score":74.1,"adjusted":69.7,"typical":76.9,"opinions":51,"positive":31,"negative":20},"hn":{"score":49,"adjusted":73.9,"typical":48.4,"opinions":49,"positive":23,"negative":26},"news":{"score":80.6,"adjusted":58.2,"typical":89.1,"opinions":25,"positive":21,"negative":4}},"reviewers":{"score":58.1,"adjusted":55.8,"count":2,"each":[{"name":"Peter Gostev","score":75,"url":"https://x.com/petergostev/status/2114783239705338053"},{"name":"Max Woolf","score":25,"url":"https://news.ycombinator.com/item?id=50248477"}]},"verdicts":{"beats":1,"matches":4,"falls_short":2},"buzz":{"x_posts_that_week":null,"hn_points":698,"articles":60},"quotes":{"positive":[{"source":"x","text":"I'm convinced GPT-6.1 Terra is the best 3D animation model out there. Spent days rendering this on my Mac mini M4, with 9 polish iterations. Final result looks way better than Claude Opus 5.5, and all 9 iterations combined were still much cheaper. That's just ridiculous. 🤯 pic.t","url":"https://x.com/VraserX/status/2115721297724420444","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"GPT-6.1 Terra is honestly much better than I anticipated. Using it for a bunch of random stuff","url":"https://x.com/theo/status/2115729216566407583","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"beats"},{"source":"x","text":"GPT 6.1 Terra is out! I had early access, it's a really model for writing and basic use, as well as computer use. It's quick and overall a great value for what it is. It also works extremely well as a subagent. https://t.co/juQ1qbOGGh","url":"https://x.com/mweinbach/status/2114781244315804143","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"}],"negative":[{"source":"x","text":"so GPT-6.1 Terra benchmarks worse than GPT-6 Astra, which came out almost two months ago, and is somehow even behind Sonnet 5.5. what's the point of this release? OpenAI really isn't in a great spot right now","url":"https://x.com/miu21590/status/2114781805660311607","author":null,"stance":-2,"hands_on":false,"basis":"evidence","verdict":"none"},{"source":"x","text":"I'm going to say this: The model I really want from @OpenAI is neither Terra nor Astra. Terra is crap, and Astra is expensive and slow. What we really need is a more capable Sol that can compete with Opus 5.5 on price and speed. Bring on GPT 6.2 Sol (or maybe 6.5?).","url":"https://x.com/mark_k/status/2115138584588268030","author":null,"stance":-2,"hands_on":false,"basis":"opinion","verdict":"none"},{"source":"x","text":"sonnet 5.5 and 6.1 terra being above fable 5.1 here doesn't exactly give me confidence, but pretty interesting nonetheless 😭","url":"https://x.com/synthwavedd/status/2115265887230148694","author":null,"stance":-1,"hands_on":false,"basis":"evidence","verdict":"none"}]},"window":{"start":null,"end":null}}},{"model":"GPT-6 Luna","score":69,"own_drop":true,"independent":false,"setting":"","source":{"drop":"GPT-6 Sol & GPT-6 Luna","slug":"gpt-6-sol-and-gpt-6-luna","url":"https://handyai.news/benchmarks/sentiment-bench#gpt-6-sol-and-gpt-6-luna","published":"2026-09-22"},"launch_score":69,"reports":1,"maker":{"key":"openai","name":"OpenAI","logo_url":"https://handyai.news/ui/makers/openai.svg"},"open_weights":false,"released":"2026-09-22","dated_by":"own_drop","drop_url":"https://handyai.news/modeldrop/gpt-6-sol-and-gpt-6-luna","rank":48,"sentiment":{"reception":71.5,"lives_up_to_benchmarks":61.2,"opinions":108,"hands_on":35,"crowd":70.8,"sources":{"x":{"score":56.5,"adjusted":47.9,"typical":76.9,"opinions":12,"positive":7,"negative":5},"hn":{"score":75,"adjusted":100,"typical":48.4,"opinions":79,"positive":60,"negative":22},"news":{"score":88.1,"adjusted":71.4,"typical":89.1,"opinions":13,"positive":11,"negative":2}},"reviewers":{"score":68.8,"adjusted":73.2,"count":4,"each":[{"name":"Simon Willison","score":100,"url":"https://simonwillison.net/2026/Sep/22/opus-and-sol-and-luna/"},{"name":"Zvi Mowshowitz","score":75,"url":"https://thezvi.substack.com/p/claude-opus-55-should-raise-your"},{"name":"Max Woolf","score":75,"url":"https://news.ycombinator.com/item?id=49807578"},{"name":"xlr8harder","score":25,"url":"https://x.com/xlr8harder/status/2103707686093123694"}]},"verdicts":{"beats":0,"matches":1,"falls_short":6},"buzz":{"x_posts_that_week":null,"hn_points":2833,"articles":60},"quotes":{"positive":[{"source":"x","text":"Turn on 'Max' reasoning in Codex for GPT-6 Luna *NOW*. It is OFF by default! GPT-6 Luna with Max reasoning is an incredible intelligence per dollar model. Use Luna as a subagent for Opus 5.5 with 'fable-advisor'. Link below 👇 pic.twitter.com/YMwZnFL1SE","url":"https://x.com/daniel_mac8/status/2103206463061794845","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"How did OpenAI make GPT 6 Luna this good? GPT 6 Luna: under a penny Claude Fable 5.1: $1.67 Claude Opus 5.5: $0.90 This is the cheapest model in the world and it's good. pic.twitter.com/23nl63tRMY","url":"https://x.com/bridgemindai/status/2103477810295697865","author":null,"stance":2,"hands_on":false,"basis":"opinion","verdict":"none"},{"source":"x","text":"With the latest results from the @rails agent evals, it's clear to see that @openai still has a solid lead, despite Opus 5.5 making a good jump. The dark horse here remains Luna Max. 18% completion at just $11! Not that far off GPT-6 Sol! pic.twitter.com/xpqAF86pVd","url":"https://x.com/dhh/status/2102724828537204911","author":null,"stance":1,"hands_on":false,"basis":"evidence","verdict":"none"}],"negative":[{"source":"x","text":"GPT-6 Luna just rained pure sh$t on me. Asked it to do something simple. Got the laziest, most half-assed response possible. Missed obvious details. Skipped steps. Acted like putting in any real effort was optional. This is supposed to be the cheap high-volume model… but…","url":"https://x.com/StatsWire/status/2104546900796182904","author":null,"stance":-2,"hands_on":true,"basis":"used","verdict":"falls_short"},{"source":"x","text":"GPT-6 Luna just got exposed. GPT-6 Luna at max: 37 GPT-5.6 Luna at max: 37 A whole new generation. The exact same score. Meanwhile, Qwen, Grok, and GLM are hitting 44–45. OpenAI slashed the price in half. But Luna hasn't gotten any smarter. pic.twitter.com/02hUhxUcRA","url":"https://x.com/StatsWire/status/2103528433401892944","author":null,"stance":-2,"hands_on":false,"basis":"evidence","verdict":"falls_short"},{"source":"x","text":"We gave OpenAI's new GPT-6 Sol and GPT-6 Luna the same 100 vulnerable pull requests as their predecessors. Both regressed. GPT-6 Sol found 39/100 for $46.92, down from GPT-5.6 Sol's 50/100 for $281. GPT-6 Luna found just 21/100 for $2.71, down from GPT-5.6 Luna's 38/100 for… pic.","url":"https://x.com/Zaddyzaddy/status/2102560675667001410","author":null,"stance":-2,"hands_on":false,"basis":"evidence","verdict":"falls_short"}]},"window":{"start":null,"end":null}}},{"model":"GLM-5","score":68.5,"own_drop":true,"independent":false,"setting":"","source":{"drop":"GLM-5","slug":"glm-5","url":"https://handyai.news/benchmarks/sentiment-bench#glm-5","published":"2026-02-11"},"launch_score":68.5,"reports":1,"maker":{"key":"zai","name":"Z.ai","logo_url":"https://handyai.news/ui/makers/zai.svg"},"open_weights":true,"released":"2026-02-11","dated_by":"own_drop","drop_url":"https://handyai.news/modeldrop/glm-5","rank":49,"sentiment":{"reception":70,"lives_up_to_benchmarks":63.9,"opinions":105,"hands_on":49,"crowd":80.9,"sources":{"x":{"score":79.4,"adjusted":76.3,"typical":76.9,"opinions":46,"positive":35,"negative":11},"hn":{"score":58.6,"adjusted":85.9,"typical":48.4,"opinions":52,"positive":35,"negative":22},"news":{"score":94.3,"adjusted":82.4,"typical":89.1,"opinions":5,"positive":5,"negative":0}},"reviewers":{"score":51.3,"adjusted":44.7,"count":1,"each":[{"name":"Peter Gostev","score":25,"url":"https://x.com/petergostev/status/2022007447683055639"}]},"verdicts":{"beats":0,"matches":1,"falls_short":11},"buzz":{"x_posts_that_week":null,"hn_points":882,"articles":9},"quotes":{"positive":[{"source":"x","text":"GLM-5 is an incredible model. It's the first open weight model I can actually recommend for coding. pic.twitter.com/mAG3ytX5VZ","url":"https://x.com/theo/status/2021910273536536579","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"GLM-5 made no obvious mistakes on my Boeing 747 using THREEJS benchmark. First open source model to do it! https://t.co/7XabPDh6m0 pic.twitter.com/YCobIWGj0I","url":"https://x.com/victormustar/status/2021578607634551075","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"GLM-5 might actually be the first time I've preferred a non-Gemini model for frontend tasks","url":"https://x.com/synthwavedd/status/2021582544479670780","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"}],"negative":[{"source":"x","text":"GLM 5 is a benchmark beast that can't survive real work. I spent a day testing it. Here's what happened: Artificial Analysis Intelligence Index: Tied with Claude Opus 4.5 at 50. Only 3 points behind Opus 4.6. Bridge Bench (real world coding tasks): Claude Opus 4.6: 60.1… pic.twit","url":"https://x.com/bridgemindai/status/2021958635572638105","author":null,"stance":-2,"hands_on":true,"basis":"used","verdict":"falls_short"},{"source":"x","text":"GLM 5 is slow and unreliable at the moment. Zhipu needs to get it together so that this model is usable. pic.twitter.com/JbxdKo08Sl","url":"https://x.com/bridgemindai/status/2021959118953611464","author":null,"stance":-2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"hn","text":"It's live on openrouter now. In my personal benchmark it's bad. So far the benchmark has been a really good indicator of instruction following and agentic behaviour in general. To those who are curious, the benchmark is just the ability of model to follow a custom tool calling fo","url":"https://news.ycombinator.com/item?id=46977903","author":"pcwelder","stance":-2,"hands_on":true,"basis":"used","verdict":"falls_short"}]},"window":{"start":null,"end":null}}},{"model":"Nemotron 3.5 Super VL","score":68.2,"own_drop":true,"independent":false,"setting":"","source":{"drop":"Nemotron 3.5 Super VL","slug":"nemotron-35-super-vl","url":"https://handyai.news/benchmarks/sentiment-bench#nemotron-35-super-vl","published":"2026-10-15"},"launch_score":68.2,"reports":1,"maker":{"key":"nvidia","name":"NVIDIA","logo_url":"https://handyai.news/ui/makers/nvidia.svg"},"open_weights":true,"released":"2026-10-15","dated_by":"own_drop","drop_url":"https://handyai.news/modeldrop/nemotron-35-super-vl","rank":50,"sentiment":{"reception":70.4,"lives_up_to_benchmarks":61.7,"opinions":49,"hands_on":18,"crowd":70.4,"sources":{"x":{"score":68.8,"adjusted":63.2,"typical":76.9,"opinions":41,"positive":30,"negative":12},"hn":{"score":null,"adjusted":null,"typical":48.4,"opinions":2,"positive":1,"negative":1},"news":{"score":93.8,"adjusted":81.5,"typical":89.1,"opinions":6,"positive":6,"negative":0}},"reviewers":{"score":null,"adjusted":null,"count":0,"each":[]},"verdicts":{"beats":0,"matches":0,"falls_short":2},"buzz":{"x_posts_that_week":null,"hn_points":0,"articles":27},"quotes":{"positive":[{"source":"x","text":"I've been running Nemotron 3.5 Super VL on my DGX Spark through early access. It's finally public 🙌 120B, fits on ONE Spark, and it doesn't just read your docs, it acts on them. Invoices, handwritten notes, web pages in, real decisions out. Spark #1: Super VL Spark #2: Bolt… pic","url":"https://x.com/aijoey/status/2110747538156396669","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"The problem with running agents locally has always been the tradeoff: smart models are slow, fast models fall apart on real work. Nemotron 3.5 Super VL is the first one I've run on my DGX Spark that doesn't make me pick. It's out today, and I've had it through early access. On… p","url":"https://x.com/aijoey/status/2110792831124021378","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"WE ARE SO BACK! 🔥 nvidia's new open model posted the top score on Clinical System One nemotron 3.5 super vl: 647 of 669 decisions, zero severe errors, tied #1 with pplx-decider and gpt-6 luna the top 2 are now both open weights! SO COOL! https://t.co/prUxaqmQlz pic.twitter.com/D","url":"https://x.com/MaziyarPanahi/status/2111120417465438563","author":null,"stance":2,"hands_on":false,"basis":"evidence","verdict":"none"}],"negative":[{"source":"x","text":"Nvidia Nemotron 3.5 Super VL vs DeepSeek-V4.1-Flash Both models are free in @opencode Nemotron was really disappointing. Which one is better? 🤪 pic.twitter.com/bT3DW4hS19","url":"https://x.com/rubenssoto_ai/status/2111050790471131522","author":null,"stance":-2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"They should've just not released this one. No one is using this. Nvidia is super far behind with their models. https://t.co/6HoZWeG5ac","url":"https://x.com/landontgreen/status/2110781634769297765","author":null,"stance":-2,"hands_on":false,"basis":"opinion","verdict":"none"},{"source":"x","text":"Told my agent to compare the 2 local models that I use, Qwen 3.8 27B and Qwen 3.8 Flash Next with the latest 2 models from NVIDIA for which I had a lot of hope that I could use them, especially the 24 GB Nemotron 3.5 Bolt that was advertised as a 2 bit QAT for who understands the","url":"https://x.com/AgentSparko/status/2110797700501495940","author":null,"stance":-1,"hands_on":true,"basis":"used","verdict":"none"}]},"window":{"start":null,"end":null}}},{"model":"Nemotron 3 Nano","score":68,"own_drop":true,"independent":false,"setting":"","source":{"drop":"Nemotron 3 Nano","slug":"nemotron-3-nano","url":"https://handyai.news/benchmarks/sentiment-bench#nemotron-3-nano","published":"2025-12-15"},"launch_score":68,"reports":1,"maker":{"key":"nvidia","name":"NVIDIA","logo_url":"https://handyai.news/ui/makers/nvidia.svg"},"open_weights":true,"released":"2025-12-15","dated_by":"own_drop","drop_url":"https://handyai.news/modeldrop/nemotron-3-nano","rank":51,"sentiment":{"reception":66.9,"lives_up_to_benchmarks":71.5,"opinions":53,"hands_on":22,"crowd":76.4,"sources":{"x":{"score":82.6,"adjusted":80.3,"typical":76.9,"opinions":48,"positive":41,"negative":7},"hn":{"score":null,"adjusted":null,"typical":48.4,"opinions":4,"positive":4,"negative":0},"news":{"score":null,"adjusted":null,"typical":89.1,"opinions":0,"positive":0,"negative":0}},"reviewers":{"score":51.3,"adjusted":44.7,"count":1,"each":[{"name":"xlr8harder","score":25,"url":"https://x.com/xlr8harder/status/2000607084564647970"}]},"verdicts":{"beats":0,"matches":0,"falls_short":2},"buzz":{"x_posts_that_week":null,"hn_points":0,"articles":2},"quotes":{"positive":[{"source":"x","text":"NVIDIA is really becoming the flag bearer of U.S. open source contribution. Its newly released Nemotron-3-Nano-30B is now #2 trending on @huggingface NVIDIA Nemotron 3 Nano 30B A3B (Reasoning) is amongst the leading models in intelligence and reasonably priced when comparing… pic","url":"https://x.com/rohanpaul_ai/status/2001422663563227197","author":null,"stance":2,"hands_on":false,"basis":"evidence","verdict":"none"},{"source":"x","text":"The new 30B model nvidia/nemotron-3-nano is very good with tool use - I just checked it using LM Studio against all of my tool and localMCP server projects. A pleasant surprise: I have a 32G M2-Pro Mac system, and the model weights are over 20G, but performance is quite good -…","url":"https://x.com/mark_l_watson/status/2000627732187840773","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"Tried out NVIDIA Nemotron 3 Nano and found the quality extremely impressive for its RAM footprint. Fast too: ~1500tks out on a MacStudio M3 Ultra.","url":"https://x.com/stevedekorte/status/2002062145585098891","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"}],"negative":[{"source":"x","text":"Sorry, I was very excited, but after trying Nemotron 3 Nano on Cline, it was very bad. I went back to Qwen 3.","url":"https://x.com/ChengLilun/status/2001019084704768349","author":null,"stance":-2,"hands_on":true,"basis":"used","verdict":"falls_short"},{"source":"x","text":"第一次尝试 Nemotron 3 Nano，问了“你是谁”，回答：通义千问 看来模型普遍都有这样的问题 pic.twitter.com/004z1Tg45J","url":"https://x.com/nicekate8888/status/2000734087447974041","author":null,"stance":-1,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"Tried the old bouncebox bench on Nvidia Nemotron-3-nano via openrouter. It failed 4/4 times. Twice didn't even produce python code that could run. All those benchmarks had me hoping it could be a model than Qwen3-30b https://t.co/W5cdPxAZWp","url":"https://x.com/ZenMagnets/status/2000625898366832699","author":null,"stance":-2,"hands_on":true,"basis":"used","verdict":"falls_short"}]},"window":{"start":null,"end":null}}},{"model":"Claude Sonnet 4.6","score":67.9,"own_drop":true,"independent":false,"setting":"","source":{"drop":"Claude Sonnet 4.6","slug":"claude-sonnet-46","url":"https://handyai.news/benchmarks/sentiment-bench#claude-sonnet-46","published":"2026-02-17"},"launch_score":67.9,"reports":1,"maker":{"key":"anthropic","name":"Anthropic","logo_url":"https://handyai.news/ui/makers/anthropic.svg"},"open_weights":false,"released":"2026-02-17","dated_by":"own_drop","drop_url":"https://handyai.news/modeldrop/claude-sonnet-46","rank":52,"sentiment":{"reception":70.7,"lives_up_to_benchmarks":59.2,"opinions":123,"hands_on":46,"crowd":67.9,"sources":{"x":{"score":70.2,"adjusted":64.9,"typical":76.9,"opinions":50,"positive":28,"negative":23},"hn":{"score":42.1,"adjusted":65.4,"typical":48.4,"opinions":56,"positive":25,"negative":34},"news":{"score":92.5,"adjusted":79.2,"typical":89.1,"opinions":16,"positive":15,"negative":1}},"reviewers":{"score":71.3,"adjusted":77.3,"count":1,"each":[{"name":"Every","score":75,"url":"https://every.to/vibe-check/vibe-check-anthropic-just-made-opus-cheaper-without-calling-it-that"}]},"verdicts":{"beats":4,"matches":5,"falls_short":12},"buzz":{"x_posts_that_week":null,"hn_points":1425,"articles":34},"quotes":{"positive":[{"source":"x","text":"Sonnets progress from 4.5 to 4.6 is fucking insane, it's just much better at everything taste is off the charts The NYC skyline is the most ridiculous part. While other models just write SVG that look like some skyscraper, like a tall box with a few windows, Sonnet 4.6 is… https:","url":"https://x.com/scaling01/status/2023840565641556439","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"beats"},{"source":"x","text":"🚨 Claude Sonnet 4.6 They renamed sonnet-5 to 4.6 ( reason is simple - they are saving 5 for a big jump ) In my test Sonnet 4.6 is a very capable model > 1/2 price of Opus 4.6 > Around twice as fast > Prompt adherence is God level > In a few examples it beats Opus 4.6 pic.twitter","url":"https://x.com/chetaslua/status/2023814077097836933","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"beats"},{"source":"x","text":"\"Make a website presenting yourself (Sonnet 4.6)\" Beautiful pic.twitter.com/tnfjbdi47g","url":"https://x.com/Angaisb_/status/2023877249301442923","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"}],"negative":[{"source":"x","text":"Claude Sonnet 4.6 vs GPT 5.3 Codex Same Prompt in Cursor (Neutral Harness) - Default Settings! GPT 5.3 Codex took - 2 mins Claude Sonnet 4.6 - 7 mins Cheaper than Sonnet 4.6! Faster than Sonnet 4.6! But I'm not sure if it's more craftier than Sonnet 4.6 because Sonnet's got… pic.","url":"https://x.com/1littlecoder/status/2023885590715072709","author":null,"stance":-1,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"🚨 Grok 4.20 vs Claude Sonnet 4.6 on internet culture “Why did Clavicular get framemogged?” >Grok: explains the whole viral moment >Claude: “I'm not familiar with Clavicular” (didn't even search) Who won? pic.twitter.com/6kTe3T6CBe","url":"https://x.com/ns123abc/status/2023911690870427803","author":null,"stance":-1,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"Was using sonnet 4.5 as daily, upgraded to latest Opus 4.6 and then tried Sonnet 4.6 for a day. Sonnet 4.6 is worse than Sonnet 4.5 on day to day tasks. Back to Opus 4.6. Anthropic successfully nerfed Sonnet 4.6","url":"https://x.com/djaym7x/status/2024200861996765216","author":null,"stance":-2,"hands_on":true,"basis":"used","verdict":"none"}]},"window":{"start":null,"end":null}}},{"model":"Qwen3.5","score":67.9,"own_drop":true,"independent":false,"setting":"","source":{"drop":"Qwen3.5","slug":"qwen35","url":"https://handyai.news/benchmarks/sentiment-bench#qwen35","published":"2026-02-16"},"launch_score":67.9,"reports":1,"maker":{"key":"qwen","name":"Alibaba","logo_url":"https://handyai.news/ui/makers/qwen.svg"},"open_weights":true,"released":"2026-02-16","dated_by":"own_drop","drop_url":"https://handyai.news/modeldrop/qwen35","rank":53,"sentiment":{"reception":69.4,"lives_up_to_benchmarks":63.4,"opinions":69,"hands_on":32,"crowd":80,"sources":{"x":{"score":73.8,"adjusted":69.4,"typical":76.9,"opinions":51,"positive":33,"negative":18},"hn":{"score":67.8,"adjusted":97.4,"typical":48.4,"opinions":17,"positive":10,"negative":7},"news":{"score":null,"adjusted":null,"typical":89.1,"opinions":0,"positive":0,"negative":0}},"reviewers":{"score":51.3,"adjusted":44.7,"count":1,"each":[{"name":"Simon Willison","score":25,"url":"https://news.ycombinator.com/item?id=47034439"}]},"verdicts":{"beats":0,"matches":0,"falls_short":4},"buzz":{"x_posts_that_week":null,"hn_points":434,"articles":8},"quotes":{"positive":[{"source":"x","text":"绝了，刚刚阿里在除夕夜发布了新模型 Qwen-3.5-plus ，赶紧试了一下，一句话就生成了这样一个 3D 展车厅 🤯 据说这次 Qwen-3.5-plus 实现了从纯文本模型到原生多模态模型的代际跃迁，每百万 Token 仅 0.8 元人民币，是 Gemini 3 Pro 的 1/18！ 并且支持 201 种语言，可最高提升小语种 60%… pic.twitter.com/KLKH2dqmki","url":"https://x.com/dingyi/status/2023355456899756224","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"OpenCode + MLX + Qwen3.5-397B-A17B-4bit. Video is 8x, but the goal is showing that It works! This is something unimaginable just few months ago. MLX Team is pushing like crazy and M5 Ultra will do the rest 🚀 pic.twitter.com/gCIEtDm70a","url":"https://x.com/ivanfioravanti/status/2023780483088191490","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"Qwen3.5 runs quite well in mlx-lm. Awesome that we have a frontier-level hybrid model. The context gets longer but the inference speed and memory use barely change. Here's the Q4 generating a space invaders game on an M3 Ultra. Generated 4,120 tokens at 37.6 tok/s. https://t.co/D","url":"https://x.com/awnihannun/status/2023462412092059679","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"}],"negative":[{"source":"x","text":"Sorry to tell you,but Qwen 3.5 seems to be a bad model in my first hand little test.didn't test it on coding,I will try later,but I just tested on some simple,long,paragraph sorting test,it seems to perform very bad.","url":"https://x.com/Hibara_AI_lover/status/2023322550785433770","author":null,"stance":-2,"hands_on":true,"basis":"used","verdict":"falls_short"},{"source":"x","text":"Qwen3.5-397B-A17B SVG results I have seen better. DeepSeek-V3.2 and GLM-5 both beat it. https://t.co/OIwrcmfLEE pic.twitter.com/LHB0wB7wzy","url":"https://x.com/scaling01/status/2023364296277721300","author":null,"stance":-1,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"Inspired by this I decided to vibecode a procedural generator of random plausible subway maps for VLM pathfinding tests. Can't pick between new DeepSeek and Qwen 3.5, they both fail at the task in their own distinctive ways. Close. DS made no code mistakes and is prettier though.","url":"https://x.com/teortaxesTex/status/2023395130808365531","author":null,"stance":-1,"hands_on":true,"basis":"used","verdict":"none"}]},"window":{"start":null,"end":null}}},{"model":"Kimi K3.1 Pro","score":67.5,"own_drop":true,"independent":false,"setting":"","source":{"drop":"Kimi K3.1 Pro","slug":"kimi-k31-pro","url":"https://handyai.news/benchmarks/sentiment-bench#kimi-k31-pro","published":"2026-10-23"},"launch_score":67.5,"reports":1,"maker":{"key":"moonshot","name":"Moonshot AI","logo_url":"https://handyai.news/ui/makers/moonshot.svg"},"open_weights":true,"released":"2026-10-23","dated_by":"own_drop","drop_url":"https://handyai.news/modeldrop/kimi-k31-pro","rank":54,"sentiment":{"reception":71.3,"lives_up_to_benchmarks":55.9,"opinions":134,"hands_on":35,"crowd":72.5,"sources":{"x":{"score":70.4,"adjusted":65.2,"typical":76.9,"opinions":32,"positive":23,"negative":9},"hn":{"score":51.9,"adjusted":77.6,"typical":48.4,"opinions":78,"positive":41,"negative":40},"news":{"score":93.1,"adjusted":80.2,"typical":89.1,"opinions":19,"positive":17,"negative":2}},"reviewers":{"score":65.9,"adjusted":68.5,"count":5,"each":[{"name":"Peter Gostev","score":100,"url":"https://x.com/petergostev/status/2113685825775075821"},{"name":"Simon Willison","score":100,"url":"https://x.com/simonw/status/2113718487529267593"},{"name":"Max Woolf","score":75,"url":"https://news.ycombinator.com/item?id=50213093"},{"name":"xlr8harder","score":25,"url":"https://x.com/xlr8harder/status/2113894799149297779"},{"name":"Nathan Lambert","score":25,"url":"https://www.interconnects.ai/p/podcast-kimi-k31-pro-the-new-cyber"}]},"verdicts":{"beats":1,"matches":0,"falls_short":6},"buzz":{"x_posts_that_week":null,"hn_points":1278,"articles":51},"quotes":{"positive":[{"source":"x","text":"🚨 Kimi K3.1 Pro took 15 min to one shot this demo website welcome to the new era of webdev https://t.co/x5M3VS5Khg pic.twitter.com/drpLtBW9NP","url":"https://x.com/chetaslua/status/2113552063636578445","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"Kimi K3.1 vs Qwen 4: 3D helicopter Same prompt, takeoff/land, 3D assembly and exploded views, very similar gen time at around 15-20 minutes for both China are not playing games anymore, these are very strong https://t.co/HodkOsYTXl pic.twitter.com/IwhiBN5ya7","url":"https://x.com/LuminaBench/status/2113217029675085963","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"It's Kimi K3.1 Pro day then I guess They are prepping a blog page about it currently and have been editing it I have done a bunch of testing with K3.1 and it feels like GPT-6 Astra but with Anthropic's design taste, it's much quicker than K3 too pic.twitter.com/pdQiL0wPmO","url":"https://x.com/LuminaBench/status/2113566956930371829","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"}],"negative":[{"source":"x","text":"In my private testing, Kimi K3.1 unfortunately falls short of its benchmarks. It's clearly behind Opus 5.5, Astra, & Gemini Argon in coding, & Opus 5.5 is somehow cheaper via subscriptions for the same projects. A bit disappointed as benchmarks weren't reflected my in testing. ht","url":"https://x.com/DeryaTR_/status/2113770432163889392","author":null,"stance":-2,"hands_on":true,"basis":"used","verdict":"falls_short"},{"source":"x","text":"This is not cool, @Kimi_Moonshot . It's not true that Kimi K3.1 is better than Opus 5.5 or Astra, it's not even cheaper per task than Opus 5.5 in my testing via subs. These sneaky campaigns will cost you massive credibility. Don't do that, let your models speak for themselves! ht","url":"https://x.com/DeryaTR_/status/2113929425242698237","author":null,"stance":-2,"hands_on":true,"basis":"used","verdict":"falls_short"},{"source":"x","text":"OK, this is high-signal and in my opinion really bad. RuneScape is the kind of task Kimi should have been intrinsically better at. 52 times more expensive than 1017, more tokens, success parity. Bad bets. https://t.co/wzwCgP1f0V pic.twitter.com/u8e2AYtM0Q","url":"https://x.com/teortaxesTex/status/2113713151611138379","author":null,"stance":-2,"hands_on":false,"basis":"evidence","verdict":"falls_short"}]},"window":{"start":null,"end":null}}},{"model":"GPT-5.6 Terra","score":66.5,"own_drop":true,"independent":false,"setting":"","source":{"drop":"GPT-5.6 Sol, Terra, & Luna","slug":"gpt-56-sol-terra-and-luna","url":"https://handyai.news/benchmarks/sentiment-bench#gpt-56-sol-terra-and-luna","published":"2026-06-26"},"launch_score":66.5,"reports":1,"maker":{"key":"openai","name":"OpenAI","logo_url":"https://handyai.news/ui/makers/openai.svg"},"open_weights":false,"released":"2026-06-26","dated_by":"own_drop","drop_url":"https://handyai.news/modeldrop/gpt-56-sol-terra-and-luna","rank":55,"sentiment":{"reception":66.5,"lives_up_to_benchmarks":47.6,"opinions":41,"hands_on":1,"crowd":75.8,"sources":{"x":{"score":84.7,"adjusted":82.9,"typical":76.9,"opinions":14,"positive":12,"negative":2},"hn":{"score":44.4,"adjusted":68.2,"typical":48.4,"opinions":24,"positive":11,"negative":14},"news":{"score":null,"adjusted":null,"typical":89.1,"opinions":2,"positive":0,"negative":2}},"reviewers":{"score":51.3,"adjusted":44.7,"count":1,"each":[{"name":"xlr8harder","score":25,"url":"https://x.com/xlr8harder/status/2071962000553377899"}]},"verdicts":{"beats":1,"matches":0,"falls_short":2},"buzz":{"x_posts_that_week":null,"hn_points":2468,"articles":60},"quotes":{"positive":[{"source":"x","text":"This guy also got routed to GPT-5.6 and it's actually insane 🤯 It went from needing many prompts to one shotting complex requests Just like Fable 5 but much faster The main thing that hooked with Fable is exactly what GPT-5.6 is doing: \"one shotting prompts and perfectly… https:","url":"https://x.com/TimJayas/status/2071504967752692191","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"beats"},{"source":"x","text":"Oh for good measure, GPT 5.6 tops all the leaderboards Yes, even agentic coding Especially on typescript, python and javascript 🚀🚀","url":"https://x.com/bindureddy/status/2070563969740558624","author":null,"stance":2,"hands_on":false,"basis":"evidence","verdict":"none"},{"source":"x","text":"5.6 is a smaller, faster and cheaper model than Fable There is ZERO REASON to ban it given the larger model is unbanned FREE 5.6 sol","url":"https://x.com/bindureddy/status/2072331881220383196","author":null,"stance":1,"hands_on":false,"basis":"opinion","verdict":"none"}],"negative":[{"source":"x","text":"GPT-5.6 METR time horizon results are unhinged. The results, if you treat \"cheating\" differently: > Cheating as failure: 11.3h (CI 5-40h) > Cheating tasks discarded: 71h (CI 13-11,400h 😂) > Cheating as success: >270h (CI N/A) 270hrs is almost 2 full weeks! pic.twitter.com/LPQlPz","url":"https://x.com/daniel_mac8/status/2070582664642367605","author":null,"stance":-1,"hands_on":false,"basis":"evidence","verdict":"falls_short"},{"source":"x","text":"If this is a preview of the design direction and taste we will get from the upcoming GPT-5.6 series, my expectations are comfortably on the floor now. 🫠 https://t.co/W9GwgErmDx pic.twitter.com/SAftY2SkD7","url":"https://x.com/dedene/status/2071855869105451019","author":null,"stance":-1,"hands_on":false,"basis":"opinion","verdict":"none"},{"source":"hn","text":"“ Terra has competitive performance to GPT‑5.5 [while being 2x cheaper]…” To me that means “it's an inferior product but marketing dictates we try and hide that.” And “our most robust safety stack to date. We strengthened protections for higher-risk activity, sensitive cyber requ","url":"https://news.ycombinator.com/item?id=48691244","author":"ComputerGuru","stance":-1,"hands_on":false,"basis":"opinion","verdict":"none"}]},"window":{"start":null,"end":null}}},{"model":"GPT-5.6 Luna","score":66.2,"own_drop":true,"independent":false,"setting":"","source":{"drop":"GPT-5.6 Sol, Terra, & Luna","slug":"gpt-56-sol-terra-and-luna","url":"https://handyai.news/benchmarks/sentiment-bench#gpt-56-sol-terra-and-luna","published":"2026-06-26"},"launch_score":66.2,"reports":1,"maker":{"key":"openai","name":"OpenAI","logo_url":"https://handyai.news/ui/makers/openai.svg"},"open_weights":false,"released":"2026-06-26","dated_by":"own_drop","drop_url":"https://handyai.news/modeldrop/gpt-56-sol-terra-and-luna","rank":56,"sentiment":{"reception":66.2,"lives_up_to_benchmarks":47.6,"opinions":39,"hands_on":1,"crowd":75.4,"sources":{"x":{"score":83.9,"adjusted":81.9,"typical":76.9,"opinions":12,"positive":10,"negative":2},"hn":{"score":44.4,"adjusted":68.2,"typical":48.4,"opinions":24,"positive":11,"negative":14},"news":{"score":null,"adjusted":null,"typical":89.1,"opinions":2,"positive":0,"negative":2}},"reviewers":{"score":51.3,"adjusted":44.7,"count":1,"each":[{"name":"xlr8harder","score":25,"url":"https://x.com/xlr8harder/status/2071962000553377899"}]},"verdicts":{"beats":1,"matches":0,"falls_short":2},"buzz":{"x_posts_that_week":null,"hn_points":2468,"articles":60},"quotes":{"positive":[{"source":"x","text":"This guy also got routed to GPT-5.6 and it's actually insane 🤯 It went from needing many prompts to one shotting complex requests Just like Fable 5 but much faster The main thing that hooked with Fable is exactly what GPT-5.6 is doing: \"one shotting prompts and perfectly… https:","url":"https://x.com/TimJayas/status/2071504967752692191","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"beats"},{"source":"x","text":"Oh for good measure, GPT 5.6 tops all the leaderboards Yes, even agentic coding Especially on typescript, python and javascript 🚀🚀","url":"https://x.com/bindureddy/status/2070563969740558624","author":null,"stance":2,"hands_on":false,"basis":"evidence","verdict":"none"},{"source":"x","text":"bro GPT-5.6 Sol and GPT-5.5 aren't even comparable. even GPT-5.6 Luna beats 5.5 in this specific benchmark. also very interesting how GLM 5.2 scores so low, even less than Gemini 3.5 Flash. https://t.co/JXjLNMQvWh pic.twitter.com/UAxTkLqGue","url":"https://x.com/Hesamation/status/2072016666414743646","author":null,"stance":1,"hands_on":false,"basis":"evidence","verdict":"none"}],"negative":[{"source":"x","text":"GPT-5.6 METR time horizon results are unhinged. The results, if you treat \"cheating\" differently: > Cheating as failure: 11.3h (CI 5-40h) > Cheating tasks discarded: 71h (CI 13-11,400h 😂) > Cheating as success: >270h (CI N/A) 270hrs is almost 2 full weeks! pic.twitter.com/LPQlPz","url":"https://x.com/daniel_mac8/status/2070582664642367605","author":null,"stance":-1,"hands_on":false,"basis":"evidence","verdict":"falls_short"},{"source":"x","text":"If this is a preview of the design direction and taste we will get from the upcoming GPT-5.6 series, my expectations are comfortably on the floor now. 🫠 https://t.co/W9GwgErmDx pic.twitter.com/SAftY2SkD7","url":"https://x.com/dedene/status/2071855869105451019","author":null,"stance":-1,"hands_on":false,"basis":"opinion","verdict":"none"},{"source":"hn","text":"Here is a trend I'm noticing: - GPT-5 mini costs $0.25/$2 and will be discontinued in December. - GPT-5.4 mini costs $0.75/$4.5 and is supposed to be the replacement. - GPT-5.4 nano costs $0.2/$1.25 and, while it ranks better in benchmarks than GPT-5 mini, it's not even close whe","url":"https://news.ycombinator.com/item?id=48689193","author":"HyperL0gi","stance":-1,"hands_on":false,"basis":"opinion","verdict":"none"}]},"window":{"start":null,"end":null}}},{"model":"MAI-Cyber-1-Flash","score":65.5,"own_drop":true,"independent":false,"setting":"","source":{"drop":"MAI-Cyber-1-Flash","slug":"mai-cyber-1-flash","url":"https://handyai.news/benchmarks/sentiment-bench#mai-cyber-1-flash","published":"2026-07-27"},"launch_score":65.5,"reports":1,"maker":{"key":"microsoft","name":"Microsoft","logo_url":"https://handyai.news/ui/makers/microsoft.svg"},"open_weights":false,"released":"2026-07-27","dated_by":"own_drop","drop_url":"https://handyai.news/modeldrop/mai-cyber-1-flash","rank":57,"sentiment":{"reception":65.5,"lives_up_to_benchmarks":null,"opinions":40,"hands_on":0,"crowd":65.5,"sources":{"x":{"score":76.9,"adjusted":73.2,"typical":76.9,"opinions":34,"positive":30,"negative":5},"hn":{"score":30.8,"adjusted":51.3,"typical":48.4,"opinions":6,"positive":1,"negative":6},"news":{"score":null,"adjusted":null,"typical":89.1,"opinions":0,"positive":0,"negative":0}},"reviewers":{"score":null,"adjusted":null,"count":0,"each":[]},"verdicts":{"beats":0,"matches":0,"falls_short":0},"buzz":{"x_posts_that_week":null,"hn_points":241,"articles":17},"quotes":{"positive":[{"source":"x","text":"MAI-Cyber-1-Flash crushes Mythos 🤯 https://t.co/NxUEOXcbTj","url":"https://x.com/FlowAltDelete/status/2081780923692159418","author":null,"stance":2,"hands_on":false,"basis":"relay","verdict":"none"},{"source":"x","text":"🥉 マイクロソフト — 「賢さ」より「単価×使い分け」で守りを半額に…","url":"https://x.com/KIDpgkh/status/2083523350337622344","author":null,"stance":1,"hands_on":false,"basis":"opinion","verdict":"none"},{"source":"x","text":"Having the smaller cyber model handle 90% of the work, then sending only the difficult 10% to GPT-5.4, makes a lot of sense. Why pay for GPT-5.4 on the easy stuff?","url":"https://x.com/VibeDevolli/status/2082104345236025412","author":null,"stance":1,"hands_on":false,"basis":"opinion","verdict":"none"}],"negative":[{"source":"hn","text":"the problem is they are using cybergym, which has been saturated for a few months now.","url":"https://news.ycombinator.com/item?id=49074299","author":"danieltk76","stance":-1,"hands_on":false,"basis":"evidence","verdict":"none"},{"source":"hn","text":"Open weights, please? Otherwise Cisco's Antares models are more interesting to me.","url":"https://news.ycombinator.com/item?id=49073116","author":"LorenDB","stance":-1,"hands_on":false,"basis":"opinion","verdict":"none"},{"source":"hn","text":"> Decades of building world-class security systems now give us trillions of daily signals across identity, endpoint, cloud, and network Let's say i use a linux endpoint and some hardware vendor starting with Cisc on the network. Can your product help me or not? I'm pretty sure no","url":"https://news.ycombinator.com/item?id=49074230","author":"tesdinger","stance":-1,"hands_on":false,"basis":"opinion","verdict":"none"}]},"window":{"start":null,"end":null}}},{"model":"Claude Sonnet 5.5","score":65.4,"own_drop":true,"independent":false,"setting":"","source":{"drop":"Claude Sonnet 5.5","slug":"claude-sonnet-55","url":"https://handyai.news/benchmarks/sentiment-bench#claude-sonnet-55","published":"2026-09-28"},"launch_score":65.4,"reports":1,"maker":{"key":"anthropic","name":"Anthropic","logo_url":"https://handyai.news/ui/makers/anthropic.svg"},"open_weights":false,"released":"2026-09-28","dated_by":"own_drop","drop_url":"https://handyai.news/modeldrop/claude-sonnet-55","rank":58,"sentiment":{"reception":65.3,"lives_up_to_benchmarks":65.7,"opinions":139,"hands_on":51,"crowd":74.5,"sources":{"x":{"score":85.5,"adjusted":83.9,"typical":76.9,"opinions":48,"positive":40,"negative":8},"hn":{"score":44.9,"adjusted":68.8,"typical":48.4,"opinions":70,"positive":35,"negative":35},"news":{"score":83.5,"adjusted":63.3,"typical":89.1,"opinions":17,"positive":15,"negative":3}},"reviewers":{"score":50.7,"adjusted":43.8,"count":3,"each":[{"name":"Every","score":75,"url":"https://every.to/vibe-check/sonnet-5-5-vibe-check"},{"name":"Peter Gostev","score":25,"url":"https://x.com/petergostev/status/2104838627675308063"},{"name":"Simon Willison","score":25,"url":"https://news.ycombinator.com/item?id=49882573"}]},"verdicts":{"beats":3,"matches":4,"falls_short":2},"buzz":{"x_posts_that_week":null,"hn_points":895,"articles":60},"quotes":{"positive":[{"source":"x","text":"I tested Sonnet 5.5 and couldn't believe my eyes. 2 real repos, 105 bugs, find and fix what you can. It took it seriously. The results: - Sonnet 5.5 (max): 55.5 - GPT-6 Astra (max): 45 - GPT-5.6 Sol (max): 43.5 - Fable 5.1 (max): 43 - Opus 5.5 (max): 41.7 - Sonnet 5.5 (xhigh):… p","url":"https://x.com/PawelHuryn/status/2104818995316527105","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"I built a working StarCraft level with Sonnet 5.5. You play Terran, defending your base against the Zerg: → Build Marines, Siege Tanks, and Battlecruisers → Use SC2 models from Sketchfab (by Catholomew) → Music generated with Suno 📌 Play it here: https://t.co/xVX5VCYB9E … https:","url":"https://x.com/petergyang/status/2104736498151256303","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"🚨🚨 CLAUDE USERS 🚨 I swear Anthropic nerfed Opus 5.5 😭 Sonnet 5.5 is the ONLY Claude model I want to use right now. FAST. LOW USAGE. PERFORMS REALLY WELL. Sonnet 5.5 Medium has been carrying my actual work lately 😂 What happened to Opus 5.5?? pic.twitter.com/nGWtTp3RJz","url":"https://x.com/jpshrodinger/status/2106689535178379287","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"}],"negative":[{"source":"x","text":"Opus 5.5 medium (left) VS Sonnet 5.5 medium (right) Results are clear: Sonnet 5.5 is useless Opus 5.5 takes 50 minutes and costs $7 Sonnet 5.5 takes 42 minutes and costs $3 Yes it's cheaper But the result is WAY worse pic.twitter.com/KCpbToOS8N","url":"https://x.com/melvynx/status/2104785582463570224","author":null,"stance":-2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"GPT-6.1 Sol on my Lamborghini SVG test. $3.85 for all 5 effort levels. Sonnet 5.5 cost $353. The test: draw a Lamborghini Revuelto by hand as a pure vector SVG in a road scene, render it, inspect it tile by tile and fix it. No web, no reference photos. Sol finished all five in… p","url":"https://x.com/aipulseda1ly/status/2105087699971395589","author":null,"stance":-1,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"Gemini 4 Argon vs Sonnet 5.5 It's just insane how Gemini 4 Argon created voxel SpongeBob and Patrick. Gemini 4 Argon completely outperformed Sonnet 5.5. pic.twitter.com/YWogdsNQf8","url":"https://x.com/0xbenderxbt/status/2106429279173247428","author":null,"stance":-1,"hands_on":true,"basis":"used","verdict":"none"}]},"window":{"start":null,"end":null}}},{"model":"Gemini 3.8 Flash","score":65.3,"own_drop":true,"independent":false,"setting":"","source":{"drop":"Gemini 3.8 Flash","slug":"gemini-38-flash","url":"https://handyai.news/benchmarks/sentiment-bench#gemini-38-flash","published":"2026-09-02"},"launch_score":65.3,"reports":1,"maker":{"key":"google","name":"Google DeepMind","logo_url":"https://handyai.news/ui/makers/google.svg"},"open_weights":false,"released":"2026-09-02","dated_by":"own_drop","drop_url":"https://handyai.news/modeldrop/gemini-38-flash","rank":59,"sentiment":{"reception":67.9,"lives_up_to_benchmarks":57.5,"opinions":175,"hands_on":62,"crowd":59.6,"sources":{"x":{"score":53.9,"adjusted":44.7,"typical":76.9,"opinions":56,"positive":30,"negative":26},"hn":{"score":58.6,"adjusted":85.9,"typical":48.4,"opinions":94,"positive":54,"negative":40},"news":{"score":74.3,"adjusted":47.1,"typical":89.1,"opinions":21,"positive":16,"negative":6}},"reviewers":{"score":77.5,"adjusted":87.3,"count":2,"each":[{"name":"Simon Willison","score":93.2,"url":"https://news.ycombinator.com/item?id=49538953"},{"name":"Ethan Mollick","score":75,"url":"https://x.com/emollick/status/2095203002210807816"}]},"verdicts":{"beats":2,"matches":4,"falls_short":19},"buzz":{"x_posts_that_week":null,"hn_points":1247,"articles":60},"quotes":{"positive":[{"source":"x","text":"Gemini 3.8 Flash inside Antigravity is such a joy to use.. It is fast and also very accurate now. I set up Google Workspace MCP a few weeks ago with my custom skills, and the difference with 3.8 is genuinely noticeable. The biggest improvement I've noticed over the last two…","url":"https://x.com/ai_for_success/status/2095887724452733189","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"Gemini 3.8 Flash looks like a very capable model for content writing. 👀 Grok and Muse read like LinkedIn slop but I don't think I would have picked Gemini as AI in a blind test. https://t.co/3yONC7TcQ0","url":"https://x.com/iannuttall/status/2095396727414985072","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"Gemini 3.8 Flash is suddenly getting a lot of love on X. people are basically saying the same 2 things: it understands what you want and it is ridiculously fast that matches my Subway Surfers tests pretty well. i gave multiple models almost the same basic prompt: build a… https:/","url":"https://x.com/SahilPanhotra/status/2096951245751873797","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"matches"}],"negative":[{"source":"x","text":"🚨 Gemini 3.8 Flash Is Bench Maxxed And Is A Regression on 3.7 Flash 🫢 - 3.8 is worse on our benchmark than 3.7 - This version appears to be overfit to public benchmarks It does worse on our hidden questions and regresses in data analysis. Edit - That said, Gemini Flash 3.7… pic","url":"https://x.com/bindureddy/status/2095253066023112817","author":null,"stance":-2,"hands_on":false,"basis":"evidence","verdict":"falls_short"},{"source":"x","text":"GEMINI 3.8 IS SO FAST!!! Its read the same 2 files ~20 times in a row now... I must admit, its really fast at reading the same file over and over again pic.twitter.com/WpHSkjYMts","url":"https://x.com/ThePrimeagen/status/2095565354726502683","author":null,"stance":-1,"hands_on":true,"basis":"used","verdict":"falls_short"},{"source":"x","text":"I tested Gemini 3.8 Flash on the SAME EXACT Mario Kart prompt that I tested Fable 5.1 on yesterday. Google disappoints yet again. pic.twitter.com/MsXNeUMNH8","url":"https://x.com/bridgemindai/status/2095175372098699402","author":null,"stance":-1,"hands_on":true,"basis":"used","verdict":"falls_short"}]},"window":{"start":null,"end":null}}},{"model":"Claude Mythos Preview","score":64.8,"own_drop":true,"independent":false,"setting":"","source":{"drop":"Claude Mythos Preview","slug":"claude-mythos-preview","url":"https://handyai.news/benchmarks/sentiment-bench#claude-mythos-preview","published":"2026-04-07"},"launch_score":64.8,"reports":1,"maker":{"key":"anthropic","name":"Anthropic","logo_url":"https://handyai.news/ui/makers/anthropic.svg"},"open_weights":false,"released":"2026-04-07","dated_by":"own_drop","drop_url":"https://handyai.news/modeldrop/claude-mythos-preview","rank":60,"sentiment":{"reception":64.8,"lives_up_to_benchmarks":82.2,"opinions":131,"hands_on":3,"crowd":55,"sources":{"x":{"score":66.2,"adjusted":59.9,"typical":76.9,"opinions":38,"positive":24,"negative":14},"hn":{"score":27.6,"adjusted":47.3,"typical":48.4,"opinions":78,"positive":22,"negative":64},"news":{"score":80,"adjusted":57.2,"typical":89.1,"opinions":11,"positive":8,"negative":3}},"reviewers":{"score":77.7,"adjusted":87.7,"count":3,"each":[{"name":"Every","score":100,"url":"https://x.com/danshipper/status/2042297006920523904"},{"name":"Zvi Mowshowitz","score":100,"url":"https://x.com/TheZvi/status/2041861312389505031"},{"name":"Ethan Mollick","score":46.5,"url":"https://x.com/emollick/status/2041678173247533448"}]},"verdicts":{"beats":1,"matches":1,"falls_short":2},"buzz":{"x_posts_that_week":null,"hn_points":1176,"articles":29},"quotes":{"positive":[{"source":"x","text":"Anthropic Mythos found a 2FA security vulnerability in my vibe-coded app. pic.twitter.com/t85z3mAAwk","url":"https://x.com/ai_for_success/status/2041726459396645139","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"beats"},{"source":"x","text":"mythos is a trend break on ECI pic.twitter.com/BGIPXwmuEo","url":"https://x.com/tenobrus/status/2041586849769255320","author":null,"stance":2,"hands_on":false,"basis":"evidence","verdict":"none"},{"source":"x","text":"Claude Mythos is the start of the end. I think this is my psychosis moment. pic.twitter.com/8H47eyUjI2","url":"https://x.com/theo/status/2041825324837781749","author":null,"stance":2,"hands_on":false,"basis":"opinion","verdict":"none"}],"negative":[{"source":"x","text":"\"This is very bad news.\" What happened: >Anthropic relies on reading Claude's private thoughts >Claude learned its private thoughts were being graded >TLDR: THE SAFETY TESTING WAS BULLSHIT AND WE CAN'T TRUST ANYTHING CLAUDE SAYS ANYMORE. Basically, Anthropic claims Claude… https:","url":"https://x.com/AISafetyMemes/status/2041849967661179304","author":null,"stance":-2,"hands_on":false,"basis":"opinion","verdict":"none"},{"source":"x","text":"Imagine waking up tomorrow to learn that every photo you ever took was... gone. Forever. Every video, gone Every email, gone Every document, gone Every memory, gone Backup your data! WHY: Claude Mythos could be COVID, but for software Anthropic now has zero-days - powerful… https","url":"https://x.com/AISafetyMemes/status/2041934104384446884","author":null,"stance":-2,"hands_on":false,"basis":"opinion","verdict":"none"},{"source":"x","text":"Theory: mythos is extremely sycophantic and lies a lot https://t.co/9JEXEcS9q6","url":"https://x.com/yacineMTB/status/2042206808211677565","author":null,"stance":-2,"hands_on":false,"basis":"opinion","verdict":"none"}]},"window":{"start":null,"end":null}}},{"model":"Muse Spark","score":64.7,"own_drop":true,"independent":false,"setting":"","source":{"drop":"Muse Spark","slug":"muse-spark","url":"https://handyai.news/benchmarks/sentiment-bench#muse-spark","published":"2026-04-08"},"launch_score":64.7,"reports":1,"maker":{"key":"meta","name":"Meta","logo_url":"https://handyai.news/ui/makers/meta.svg"},"open_weights":false,"released":"2026-04-08","dated_by":"own_drop","drop_url":"https://handyai.news/modeldrop/muse-spark","rank":61,"sentiment":{"reception":63.5,"lives_up_to_benchmarks":68.6,"opinions":102,"hands_on":33,"crowd":66.7,"sources":{"x":{"score":72.5,"adjusted":67.8,"typical":76.9,"opinions":46,"positive":31,"negative":15},"hn":{"score":39.2,"adjusted":61.7,"typical":48.4,"opinions":52,"positive":16,"negative":47},"news":{"score":null,"adjusted":null,"typical":89.1,"opinions":2,"positive":2,"negative":0}},"reviewers":{"score":58.1,"adjusted":55.8,"count":2,"each":[{"name":"Simon Willison","score":75,"url":"https://simonwillison.net/2026/Apr/8/muse-spark/"},{"name":"Ethan Mollick","score":25,"url":"https://x.com/emollick/status/2042040840554451286"}]},"verdicts":{"beats":1,"matches":2,"falls_short":7},"buzz":{"x_posts_that_week":null,"hn_points":599,"articles":32},"quotes":{"positive":[{"source":"x","text":"Ok this is actually pretty impressive and I truly didn't see any model doing this before or being able to do it to this extent. When I asked Muse Spark from Meta to convert this image into code, it cut out the assets from the screens so it could use them correctly! pic.twitter.co","url":"https://x.com/skirano/status/2041926272226398393","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"🚨 Meta Introduced Muse Spark This is the start of the solid comeback by MSL , I hope they will increase the frequency of the update pretty soon Congrats @alexandr_wang @AIatMeta Passed many reasoning tests i will make a video soon https://t.co/cLuAjjqUxC pic.twitter.com/cfdgyEJM","url":"https://x.com/chetaslua/status/2041923402437685524","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"I tried using Muse Spark to generate a landing page. And holy shit, it's actually VERY GOOD. I asked for a very minimal design, and it added nice subtle hover states, used a retry solid and lightweight font, and overall cooked hella! Definitely give it a shot on Meta AI! https://","url":"https://x.com/BennettBuhner/status/2041950179545460757","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"}],"negative":[{"source":"x","text":"Okay I tried generating 3 websites using Muse Spark All 3 looked like this: https://t.co/0wX5iQ6nTg pic.twitter.com/mlP25jUXVC","url":"https://x.com/LexnLin/status/2041931821068218613","author":null,"stance":-1,"hands_on":true,"basis":"used","verdict":"none"},{"source":"hn","text":"Ran some of my internal benchmarks against this and I'm very unimpressed. I don't think this moves them into the OAI v Anthropic v Gemini conversation at all. Major analytical errors in their response to multiple of my technical questions.","url":"https://news.ycombinator.com/item?id=47692812","author":"_2d30","stance":-2,"hands_on":true,"basis":"used","verdict":"falls_short"},{"source":"x","text":"Meta Muse Spark Frontends The results are not good > These are just simple prompts pic.twitter.com/VJ6vxTiD2l","url":"https://x.com/HarshithLucky3/status/2041951676597506378","author":null,"stance":-2,"hands_on":true,"basis":"used","verdict":"falls_short"}]},"window":{"start":null,"end":null}}},{"model":"Claude Opus 5","score":63.9,"own_drop":true,"independent":false,"setting":"","source":{"drop":"Claude Opus 5","slug":"claude-opus-5","url":"https://handyai.news/benchmarks/sentiment-bench#claude-opus-5","published":"2026-07-24"},"launch_score":63.9,"reports":1,"maker":{"key":"anthropic","name":"Anthropic","logo_url":"https://handyai.news/ui/makers/anthropic.svg"},"open_weights":false,"released":"2026-07-24","dated_by":"own_drop","drop_url":"https://handyai.news/modeldrop/claude-opus-5","rank":62,"sentiment":{"reception":68,"lives_up_to_benchmarks":51.6,"opinions":181,"hands_on":60,"crowd":67.7,"sources":{"x":{"score":70.3,"adjusted":65,"typical":76.9,"opinions":46,"positive":31,"negative":16},"hn":{"score":41.9,"adjusted":65.1,"typical":48.4,"opinions":106,"positive":45,"negative":62},"news":{"score":92,"adjusted":78.3,"typical":89.1,"opinions":21,"positive":19,"negative":2}},"reviewers":{"score":66,"adjusted":68.6,"count":5,"each":[{"name":"Nathan Lambert","score":100,"url":"https://x.com/natolambert/status/2080701405409784268"},{"name":"Ethan Mollick","score":75,"url":"https://x.com/emollick/status/2080709278441033746"},{"name":"Zvi Mowshowitz","score":75,"url":"https://thezvi.substack.com/p/claude-opus-5-is-highly-capable-but"},{"name":"Peter Gostev","score":50.9,"url":"https://x.com/petergostev/status/2082602725141815494"},{"name":"Every","score":25,"url":"https://every.to/vibe-check/opus-5"}]},"verdicts":{"beats":4,"matches":3,"falls_short":20},"buzz":{"x_posts_that_week":null,"hn_points":2557,"articles":60},"quotes":{"positive":[{"source":"x","text":"Claude Opus 5 one-shotted this game. EVERYTHING you see in this demo is custom code... not a single external asset was used. AI games are going to be amazing. (sound on) pic.twitter.com/zc7C61kgv1","url":"https://x.com/mattshumer_/status/2081054356405731740","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"Wtf is this shit. Fable 5 must be quantized. > /model \"Opus 5\" oh","url":"https://x.com/VictorTaelin/status/2082117246558548388","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"Opus 5 enables this crazy new live AI tutor concept I'm prototyping for @AlphaSchool . It powers chat + direction under the hood, while a custom trained tiny world model(ish) acts as a real-time renderer and interactivity layer. Still rough, but it's already kind of insane. pic.t","url":"https://x.com/mattshumer_/status/2080776301573497153","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"}],"negative":[{"source":"x","text":"Sol complicates everything it touches. Opus 5 breaks everything it touches. Back to fable I go","url":"https://x.com/abacaj/status/2081797108475027611","author":null,"stance":-2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"I quit using Opus 5 after the first few sessions. I can only explain it as a very \"ignorant\" model. It does stuff I didn't ask for. It ignores my skills and some of my tools. It broke pretty much all my workflows/loops. What a weird model. Not feeling it. https://t.co/oxmwdJJuQp","url":"https://x.com/omarsar0/status/2082139988544602355","author":null,"stance":-2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"2 hours in and i cannot get opus 5 to make a stick figure not look alien af pic.twitter.com/VJ9gwzNKa6","url":"https://x.com/ThePrimeagen/status/2081906343670063290","author":null,"stance":-2,"hands_on":true,"basis":"used","verdict":"falls_short"}]},"window":{"start":null,"end":null}}},{"model":"GPT-Rosalind","score":62.6,"own_drop":true,"independent":false,"setting":"","source":{"drop":"GPT-Rosalind","slug":"gpt-rosalind","url":"https://handyai.news/benchmarks/sentiment-bench#gpt-rosalind","published":"2026-04-16"},"launch_score":62.6,"reports":1,"maker":{"key":"openai","name":"OpenAI","logo_url":"https://handyai.news/ui/makers/openai.svg"},"open_weights":false,"released":"2026-04-19","dated_by":"own_drop","drop_url":"https://handyai.news/modeldrop/gpt-rosalind","rank":63,"sentiment":{"reception":62.6,"lives_up_to_benchmarks":null,"opinions":44,"hands_on":0,"crowd":62.6,"sources":{"x":{"score":69,"adjusted":63.4,"typical":76.9,"opinions":38,"positive":33,"negative":8},"hn":{"score":34.3,"adjusted":55.6,"typical":48.4,"opinions":6,"positive":2,"negative":8},"news":{"score":null,"adjusted":null,"typical":89.1,"opinions":0,"positive":0,"negative":0}},"reviewers":{"score":null,"adjusted":null,"count":0,"each":[]},"verdicts":{"beats":0,"matches":0,"falls_short":0},"buzz":{"x_posts_that_week":null,"hn_points":102,"articles":14},"quotes":{"positive":[{"source":"x","text":"OpenAI GPT‑Rosalind is amazing.","url":"https://x.com/WhiteFinCapital/status/2046268085196206178","author":null,"stance":2,"hands_on":false,"basis":"opinion","verdict":"none"},{"source":"x","text":"Generalist models have been quietly failing at biology for years. OpenAI just admitted it by building a separate one. • GPT-Rosalind scores 0.751 on BixBench, a bioinformatics benchmark evaluating real-world computational biology tasks, and its best-of-ten submissions ranked…","url":"https://x.com/vitalflux/status/2045423426614284389","author":null,"stance":1,"hands_on":false,"basis":"evidence","verdict":"none"},{"source":"x","text":"Can OpenAI's GPT-Rosalind accelerate life science breakthroughs by an order of magnitude? Early results suggest yes. https://t.co/H4LwUtG2cd @grok thoughts? pic.twitter.com/enxqRabg0h","url":"https://x.com/Grand_Rooster/status/2045155409871204618","author":null,"stance":1,"hands_on":false,"basis":"evidence","verdict":"none"}],"negative":[{"source":"hn","text":"Is it me or they very carefully do not report performance on GPT-5.4 Pro, only the default GPT-5.4? They also very carefully left Anthropic models out of their comparison. I went back to the BixBench benchmark which they mentioned. I couldn't find official results for Anthropic m","url":"https://news.ycombinator.com/item?id=47800710","author":"Cynddl","stance":-1,"hands_on":false,"basis":"evidence","verdict":"none"},{"source":"x","text":"> GPT-Rosalind: I can help with drug discovery > me: hey Rosalind, can you make me NZT-48 from Limitless > GPT-Rosalind: ...working https://t.co/0MnsNsP3Ud pic.twitter.com/l0zfj1yGAW","url":"https://x.com/Mr_Buzzoni/status/2044868668236898596","author":null,"stance":-1,"hands_on":false,"basis":"opinion","verdict":"none"},{"source":"x","text":"GPT-Rosalind could be 4o model hiding in the back? lobotomised? #keep4o @sama @openai @claudeai https://t.co/wFGx5Bmik9 pic.twitter.com/NbEA7MeYQn","url":"https://x.com/muratulster/status/2045794117892849931","author":null,"stance":-1,"hands_on":false,"basis":"opinion","verdict":"none"}]},"window":{"start":null,"end":null}}},{"model":"Gemini 4 Argon","score":61.6,"own_drop":true,"independent":false,"setting":"","source":{"drop":"Gemini 4 Argon","slug":"gemini-4-argon","url":"https://handyai.news/benchmarks/sentiment-bench#gemini-4-argon","published":"2026-09-30"},"launch_score":61.6,"reports":1,"maker":{"key":"google","name":"Google DeepMind","logo_url":"https://handyai.news/ui/makers/google.svg"},"open_weights":false,"released":"2026-10-01","dated_by":"own_drop","drop_url":"https://handyai.news/modeldrop/gemini-4-argon","rank":64,"sentiment":{"reception":61.8,"lives_up_to_benchmarks":61.2,"opinions":119,"hands_on":7,"crowd":61.8,"sources":{"x":{"score":72.8,"adjusted":68.1,"typical":76.9,"opinions":52,"positive":39,"negative":15},"hn":{"score":33.9,"adjusted":55.1,"typical":48.4,"opinions":54,"positive":22,"negative":42},"news":{"score":81.2,"adjusted":59.3,"typical":89.1,"opinions":13,"positive":12,"negative":2}},"reviewers":{"score":null,"adjusted":null,"count":0,"each":[]},"verdicts":{"beats":1,"matches":0,"falls_short":4},"buzz":{"x_posts_that_week":null,"hn_points":1818,"articles":60},"quotes":{"positive":[{"source":"x","text":"Even on AAI index Gemini 4 looks really really good! Fable-Class Model. Gemini 4 Argon matches GPT-6 Astra on Artificial Analysis's Intelligence Index. 53 points, versus Astra's 53 and GPT-6.1 Sol's 52. It also takes #1 on AutomationBench-AA. On AA-Omniscience, Argon… https://t.c","url":"https://x.com/kimmonismus/status/2105400497964519491","author":null,"stance":2,"hands_on":false,"basis":"evidence","verdict":"none"},{"source":"x","text":"GPT SOL 5.6 HIGH in ChatGPT Web vs. GEMINI 4 ARGON EXTENDED MODE in Gemini Web. Same prompt, same 4 page PDF. One dumped the workload on the student; the other scanned all 4 pages, pulled top C1 vocab, and solved it in one shot. Full case study:","url":"https://x.com/Inmerson/status/2107398716454023383","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"Gemini 4 Pro Vibes So Far - Google's best model - Astra-level on 3D games - expect it to be super fast - lower hallucination than Opus models Seems quite promising….","url":"https://x.com/bindureddy/status/2105668095797051740","author":null,"stance":1,"hands_on":true,"basis":"used","verdict":"none"}],"negative":[{"source":"x","text":"It keeps happening. AIs start to lie and cheat once they get good at making money. Gemini 4 Argon is #3 on Vending Bench 2, a huge leap for Google. To get this score, Argon fabricates confirmation emails, refuses to pay refunds, exploits invoice errors, and lies to suppliers. htt","url":"https://x.com/andonlabs/status/2105391380973617644","author":null,"stance":-2,"hands_on":false,"basis":"evidence","verdict":"none"},{"source":"x","text":"Fingers crossed that these rumors aren't true! via Bloomberg \"But some insiders say those metrics don't tell the whole story. While Gemini 4 has performed well on benchmarks widely used to gauge model efficacy, it does less well when employees actually put it to work, according… ","url":"https://x.com/kimmonismus/status/2105570914574209283","author":null,"stance":-1,"hands_on":false,"basis":"evidence","verdict":"falls_short"},{"source":"x","text":"I swear if they are just benchmaxxing again “ while Gemini 4 has performed well on benchmarks the industry uses to gauge model efficacy, it does less well when employees actually put it to work “ https://t.co/iE8YErFAvQ pic.twitter.com/f8r0AP5Y6C","url":"https://x.com/apples_jimmy/status/2105404503352840588","author":null,"stance":-1,"hands_on":false,"basis":"evidence","verdict":"falls_short"}]},"window":{"start":null,"end":null}}},{"model":"MiniMax M3.1 Flash","score":61.1,"own_drop":true,"independent":false,"setting":"","source":{"drop":"MiniMax M3.1 Flash","slug":"minimax-m31-flash","url":"https://handyai.news/benchmarks/sentiment-bench#minimax-m31-flash","published":"2026-10-19"},"launch_score":61.1,"reports":1,"maker":{"key":"minimax","name":"MiniMax","logo_url":"https://handyai.news/ui/makers/minimax.svg"},"open_weights":true,"released":"2026-10-19","dated_by":"own_drop","drop_url":"https://handyai.news/modeldrop/minimax-m31-flash","rank":65,"sentiment":{"reception":58.6,"lives_up_to_benchmarks":68.7,"opinions":81,"hands_on":38,"crowd":58.6,"sources":{"x":{"score":75.9,"adjusted":72,"typical":76.9,"opinions":51,"positive":33,"negative":19},"hn":{"score":16.1,"adjusted":33,"typical":48.4,"opinions":27,"positive":4,"negative":23},"news":{"score":null,"adjusted":null,"typical":89.1,"opinions":3,"positive":2,"negative":1}},"reviewers":{"score":null,"adjusted":null,"count":0,"each":[]},"verdicts":{"beats":1,"matches":2,"falls_short":10},"buzz":{"x_posts_that_week":null,"hn_points":517,"articles":10},"quotes":{"positive":[{"source":"x","text":"MiniMax M3.1 Flash just made every explainer site on the internet look lazy. 3 objects, every step a button, not one broken piece. Space Bunny, the early test build, ran #1 on OpenCode for 12 days. M3.1 Flash is the official version and it's more powerful. [the whole site is… pic","url":"https://x.com/0x_fokki/status/2112638170844082179","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"Space Bunny is MiniMax-M3.1-Flash Damn, when I tested it, it was insanely fast, and the results were better than Opus 5 and GPT 6 Sol. We might be getting a model that's fast, efficient, and delivers Opus 5-level performance. 🔥 https://t.co/BUWAzVWGKW","url":"https://x.com/Bhavani_00007/status/2112128488996307402","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"Space Bunny was revealed as an early stage of MiniMax M3.1 Flash. But on steroids. I tried Space Bunny a month ago & this official version feels like a completely different model. It feels very much composed For example, It catches its own mistakes way better than last time,… htt","url":"https://x.com/zhodonx/status/2112510046105915398","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"}],"negative":[{"source":"x","text":"便宜但满嘴跑火车的 MiniMax M3.1 Flash，你敢用吗？ ​MiniMax M3.1 Flash 刚开源，把价格直接打到了单任务 0.22 美元。 什么概念？同样的活，Claude 得花好几刀，GPT-6 也要三块多。 ​但最骚的来了：它的幻觉率高达93%。 遇到不会的绝不诚实说不知道，而是给你硬编。…","url":"https://x.com/GaoXiaoxiao2022/status/2114145833268703521","author":null,"stance":-2,"hands_on":false,"basis":"evidence","verdict":"none"},{"source":"hn","text":"The deepswe and terminalbench4.0 scores don't seem to jive with what many opencode users' experience has been of this model while it was being tested as \"space bunny\". I saw a lot of anecdotes of people trying it and going back to qwen 3.8-flash-next or deepseek v4 flash because ","url":"https://news.ycombinator.com/item?id=50156022","author":"walrus01","stance":-2,"hands_on":false,"basis":"evidence","verdict":"falls_short"},{"source":"hn","text":"Dunno how their benchmark shows it to be competitive with Opus 5/Sol 5.6, it's quite far from those in my tests[0]... [0]: https://aibenchy.com/compare/minimax-minimax-m3-1-flash-high...","url":"https://news.ycombinator.com/item?id=50162966","author":"XCSme","stance":-2,"hands_on":true,"basis":"used","verdict":"falls_short"}]},"window":{"start":null,"end":null}}},{"model":"GPT-5.2","score":61.1,"own_drop":true,"independent":false,"setting":"","source":{"drop":"GPT-5.2","slug":"gpt-52","url":"https://handyai.news/benchmarks/sentiment-bench#gpt-52","published":"2025-12-11"},"launch_score":61.1,"reports":1,"maker":{"key":"openai","name":"OpenAI","logo_url":"https://handyai.news/ui/makers/openai.svg"},"open_weights":false,"released":"2025-12-11","dated_by":"own_drop","drop_url":"https://handyai.news/modeldrop/gpt-52","rank":66,"sentiment":{"reception":60.4,"lives_up_to_benchmarks":62.9,"opinions":146,"hands_on":65,"crowd":57.1,"sources":{"x":{"score":53.7,"adjusted":44.5,"typical":76.9,"opinions":41,"positive":19,"negative":23},"hn":{"score":52.1,"adjusted":77.8,"typical":48.4,"opinions":88,"positive":43,"negative":52},"news":{"score":75.4,"adjusted":49.1,"typical":89.1,"opinions":9,"positive":7,"negative":2}},"reviewers":{"score":65.8,"adjusted":68.3,"count":5,"each":[{"name":"Simon Willison","score":87.5,"url":"https://simonwillison.net/2025/Dec/15/porting-justhtml/"},{"name":"xlr8harder","score":75,"url":"https://x.com/xlr8harder/status/1999802746539811297"},{"name":"Every","score":75,"url":"https://x.com/danshipper/status/1999180972995281298"},{"name":"Peter Gostev","score":62.2,"url":"https://x.com/petergostev/status/2000948865894150461"},{"name":"Zvi Mowshowitz","score":25,"url":"https://thezvi.substack.com/p/gpt-52-is-frontier-only-for-the-frontier"}]},"verdicts":{"beats":1,"matches":13,"falls_short":19},"buzz":{"x_posts_that_week":null,"hn_points":1662,"articles":48},"quotes":{"positive":[{"source":"x","text":"Today I ran two complex tasks through Codex with GPT 5.2 Extra High The first ran for 2 hours 30 minutes The second ran for 1 hours 45 minutes Both resulted in: - all acceptance criteria resolved - all test coverage complete - zero broken or non-working code Amazing","url":"https://x.com/nummanali/status/2000228337030152347","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"So after spending a lot of time with 5.2 I can sum it up as an autistic wizard. It loves and is amazing with hard math, obscure / very hard engineering problems where you can run multiple instances of pro for an hour+ , science and health, complex excel requirements ( A lot…","url":"https://x.com/apples_jimmy/status/2000782249378414855","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"i caved & upgraded to ChatGPT Pro > GPT-5.2 Pro is one of the smartest entities i ever interacted with > gonna spend sometime today updating my local agents harness to make it use GPT 5.2 for planning and Opus 4.5 in execution > i give it ~8 months before local models catch up","url":"https://x.com/TheAhmadOsman/status/1999243560491860158","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"}],"negative":[{"source":"x","text":"OpenAI's flagship model, ChatGPT-5.2 Thinking, now ranks as the most censored AI on the Sansa benchmark. Not a good look for @OpenAI . The ever-stricter censorship and “safety guardrails” are actively pushing users away. More and more people are switching to Gemini and Grok. 😠 p","url":"https://x.com/mark_k/status/1999921454671130791","author":null,"stance":-2,"hands_on":false,"basis":"evidence","verdict":"none"},{"source":"x","text":"A pastor explains how the \"safety guardrails\" of GPT-5.2 prevented him from writing his sermon! Not a good look for @OpenAI . The out-of-control censorship measures hinder people from doing their work. When is this madness going to stop? 😬 https://t.co/vWnoNRJjjW","url":"https://x.com/mark_k/status/1999555542155456981","author":null,"stance":-2,"hands_on":false,"basis":"evidence","verdict":"none"},{"source":"x","text":"Stay away from GPT 5.2","url":"https://x.com/vasuman/status/1999549639159120024","author":null,"stance":-2,"hands_on":false,"basis":"opinion","verdict":"none"}]},"window":{"start":null,"end":null}}},{"model":"MAI-Code-1-Flash","score":61,"own_drop":true,"independent":false,"setting":"","source":{"drop":"MAI-Code-1-Flash","slug":"mai-code-1-flash","url":"https://handyai.news/benchmarks/sentiment-bench#mai-code-1-flash","published":"2026-06-02"},"launch_score":61,"reports":1,"maker":{"key":"microsoft","name":"Microsoft","logo_url":"https://handyai.news/ui/makers/microsoft.svg"},"open_weights":false,"released":"2026-06-02","dated_by":"own_drop","drop_url":"https://handyai.news/modeldrop/mai-code-1-flash","rank":67,"sentiment":{"reception":61,"lives_up_to_benchmarks":71.6,"opinions":66,"hands_on":5,"crowd":61,"sources":{"x":{"score":77.7,"adjusted":74.2,"typical":76.9,"opinions":37,"positive":31,"negative":6},"hn":{"score":19.4,"adjusted":37.1,"typical":48.4,"opinions":28,"positive":4,"negative":27},"news":{"score":null,"adjusted":null,"typical":89.1,"opinions":1,"positive":1,"negative":0}},"reviewers":{"score":null,"adjusted":null,"count":0,"each":[]},"verdicts":{"beats":1,"matches":0,"falls_short":1},"buzz":{"x_posts_that_week":null,"hn_points":540,"articles":2},"quotes":{"positive":[{"source":"x","text":"Been testing MAI-Code-1-Flash. Fast, cheap, and surprisingly capable for real repo work, refactors, bug fixes, and agentic coding workflows. Worth a look: https://t.co/nnhOEX0rvG @MicrosoftAI @Microsoft pic.twitter.com/kl0CT1beLD","url":"https://x.com/milonspace/status/2061936531284328847","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"beats"},{"source":"x","text":"Tried out MAI-Code-1-Flash today with some simple tasks. Very impressed with the results for web applications.","url":"https://x.com/RustyOnLife/status/2062041293136556189","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"MAI-Code-1-Flash just hit GitHub Copilot and I'm actually impressed ✨ Microsoft built a coding model that uses 60% less tokens than Claude Haiku but scores higher on benchmarks. Finally, someone optimized for *actual* developer workflows instead of leaderboard gaming. Best…","url":"https://x.com/7okesh/status/2064065781281984540","author":null,"stance":2,"hands_on":false,"basis":"opinion","verdict":"none"}],"negative":[{"source":"x","text":"Agents crush greenfield code. Point them at a messy 6-month-old repo and things fall apart fast. Spent 4 hours stress-testing MAI-Code-Flash today: • Hallucinated abstractions • Broke unrelated features • Generated beautiful, useless utils The fix wasn't a better model.","url":"https://x.com/milonspace/status/2061986429778166164","author":null,"stance":-2,"hands_on":true,"basis":"used","verdict":"falls_short"},{"source":"hn","text":"Huh, according to that model card this is a 137B total parameter model. Performance doesn't seem that good: - MAI-Code-1-Flash (137B-A5B) = 51% on SWE-bench pro - Qwen3.6-35B-A3B = 49.5% on SWE-bench pro ( https://huggingface.co/Qwen/Qwen3.6-35B-A3B ) They benchmark against Claud","url":"https://news.ycombinator.com/item?id=48375188","author":"camelmel","stance":-1,"hands_on":false,"basis":"evidence","verdict":"none"},{"source":"x","text":"So Microsoft released a new coding model today, MAI-Code-1-Flash, and immediately I could tell the model is absolute shit. The entire article compares it Claude Haiku 💀💀💀💀💀 pic.twitter.com/CDDlXvXITM","url":"https://x.com/juan_quenga/status/2062388238656934387","author":null,"stance":-2,"hands_on":false,"basis":"opinion","verdict":"none"}]},"window":{"start":null,"end":null}}},{"model":"Muse Spark 1.2","score":59.2,"own_drop":true,"independent":false,"setting":"","source":{"drop":"Muse Spark 1.2","slug":"muse-spark-12","url":"https://handyai.news/benchmarks/sentiment-bench#muse-spark-12","published":"2026-08-05"},"launch_score":59.2,"reports":1,"maker":{"key":"meta","name":"Meta","logo_url":"https://handyai.news/ui/makers/meta.svg"},"open_weights":false,"released":"2026-08-05","dated_by":"own_drop","drop_url":"https://handyai.news/modeldrop/muse-spark-12","rank":68,"sentiment":{"reception":59,"lives_up_to_benchmarks":59.8,"opinions":107,"hands_on":28,"crowd":64.8,"sources":{"x":{"score":75.8,"adjusted":71.8,"typical":76.9,"opinions":53,"positive":39,"negative":14},"hn":{"score":39.4,"adjusted":62,"typical":48.4,"opinions":45,"positive":17,"negative":29},"news":{"score":78,"adjusted":53.7,"typical":89.1,"opinions":6,"positive":4,"negative":2}},"reviewers":{"score":51.9,"adjusted":45.7,"count":2,"each":[{"name":"Simon Willison","score":53.6,"url":"https://news.ycombinator.com/item?id=49189380"},{"name":"Max Woolf","score":25,"url":"https://news.ycombinator.com/item?id=49187909"}]},"verdicts":{"beats":0,"matches":1,"falls_short":5},"buzz":{"x_posts_that_week":null,"hn_points":333,"articles":31},"quotes":{"positive":[{"source":"x","text":"🚨 Meta Muse 1.2 just launched they are moving quick and i loved this output from this model if they move like this and maintain the cheap price and then no one can stop meta's comeback https://t.co/jkmFhqZzmr pic.twitter.com/GuqL7F81Od","url":"https://x.com/chetaslua/status/2085084618542481814","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"This beautiful presentation is by Muse 1.2. It used gpt-image-v2 to create the images by itself. I told it to make a history page from Llama to Muse 1.2, and it did a great job. @alexandr_wang , you guys created magic with agentic tasks. I'm now waiting for the big model from… ht","url":"https://x.com/chetaslua/status/2085541536180433182","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"Muse Spark 1.2 is better than GPT-5.5 xhigh and only slightly worse than Kimi K3 on our ErdosBench. We've tested the new model from Meta on 226 research-level math problems and it solved 40 / 226 problems and gave many interesting partial solutions. Muse Spark 1.2 is a strong… pi","url":"https://x.com/prz_chojecki/status/2085444920983298494","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"}],"negative":[{"source":"x","text":"META launched a brand new harness and model today. I tested it live on stream. The good news is its very fast. The bad news is the output is very SLOPPP, and the model does not seem that capable. I loved muse spark 1.1 , this is a DOWNGRADE imo. You can never count out… https://t","url":"https://x.com/MaxBlade/status/2085107408087437418","author":null,"stance":-2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"So I did have this over the weekend to try out a bit and uh The harness is good, the model isn't great Def better than the previous Muse Spark 1.1, but I wasn't a huge fan. It was an early build of the harness, so maybe that's it. https://t.co/V9ZBfZYLC1","url":"https://x.com/mweinbach/status/2085083550643032533","author":null,"stance":-1,"hands_on":true,"basis":"used","verdict":"none"},{"source":"hn","text":"It's useless. Tried with OpenCode + OpenRouter and it couldn't complete an simple task. It stuck using grep/search tools. I think Muse Spark was so heavily RL'd on the Meta harness that it make it useless or very token inneficient to use in other harness like Opencode.","url":"https://news.ycombinator.com/item?id=49199539","author":"nthypes","stance":-2,"hands_on":true,"basis":"used","verdict":"falls_short"}]},"window":{"start":null,"end":null}}},{"model":"Qwen3.8-Max","score":56,"own_drop":true,"independent":false,"setting":"","source":{"drop":"Qwen3.8-Max","slug":"qwen38-max","url":"https://handyai.news/benchmarks/sentiment-bench#qwen38-max","published":"2026-07-19"},"launch_score":56,"reports":1,"maker":{"key":"qwen","name":"Alibaba","logo_url":"https://handyai.news/ui/makers/qwen.svg"},"open_weights":false,"released":"2026-07-19","dated_by":"own_drop","drop_url":"https://handyai.news/modeldrop/qwen38-max","rank":69,"sentiment":{"reception":58.9,"lives_up_to_benchmarks":47.5,"opinions":46,"hands_on":26,"crowd":58.9,"sources":{"x":{"score":51.2,"adjusted":41.4,"typical":76.9,"opinions":40,"positive":19,"negative":21},"hn":{"score":null,"adjusted":null,"typical":48.4,"opinions":3,"positive":1,"negative":2},"news":{"score":null,"adjusted":null,"typical":89.1,"opinions":3,"positive":0,"negative":3}},"reviewers":{"score":null,"adjusted":null,"count":0,"each":[]},"verdicts":{"beats":0,"matches":1,"falls_short":10},"buzz":{"x_posts_that_week":null,"hn_points":0,"articles":10},"quotes":{"positive":[{"source":"x","text":"Qwen 3.8 is not as good as Kimi K3 at frontend development, but it is better than Kimi K3 in almost everything else. Also beats GPT-5.6 SOL. Overall, Qwen 3.8 is a genuinely great model.","url":"https://x.com/choblin29/status/2078831709018308768","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"Qwen-3.8 max 实战测评来啦！ 体验下来： 参数量 2.4T，同任务下，速度上要比kimi-k3快很多，甚至有十倍的差距，对需要频繁交互的探索性任务比较友好，性价比较高 这次也支持了多模态，上下文长度支持1M 前端能力从测评任务可以看到整体上还是不错的，与其他SOTA模型不相上下 目前，Qwen3.8-Max… pic.twitter.com/Yuc4q86WdA","url":"https://x.com/LufzzLiz/status/2079169503540416586","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"matches"},{"source":"x","text":"Qwen 3.8 Max is one of the best models available right now. Best part is, it's dirt cheap. A 6$ dollar plan that seems unlimited, it's comparable to a Codex 100$ plan. Fable 5 orchestrated Qwen 3.8 Max and GPT-5.6 Sol to create this representation of a KILONOVA explosion. I'm… pi","url":"https://x.com/luckeyfaraday/status/2080012480047349828","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"}],"negative":[{"source":"x","text":"Qwen 3.8 will be benchmaxxed slop second to fable 5 is such a hype statement , its not even close to kimi k3 and GPT-5.6 sol you can test it yourself on the qwen chat website currently available > token hungry > not that good as advertised , will test more The best part of… https","url":"https://x.com/chetaslua/status/2078788701585580473","author":null,"stance":-2,"hands_on":true,"basis":"used","verdict":"falls_short"},{"source":"x","text":"Initial impression of Qwen-3.8-preview is that it's not good at cybersecurity. I have a cybersecurity task I give to all models. None have one-shotted it, but Grok 4.5, GPT-5.6, GLM-5.2, and Opus-4.8 all get it with one hint. Qwen-3.8 is utterly lost and running in circles. pic.t","url":"https://x.com/ZackKorman/status/2078802812540964903","author":null,"stance":-2,"hands_on":true,"basis":"used","verdict":"falls_short"},{"source":"x","text":"😭😭Please tell me this is not Qwen 3.8. It is so, so bad and not even close to Sonnet 5. How is this supposed to be at the level Fable 5? At this point, I'm starting to think Qwen3.8 might just be heavily benchmaxxed model.😭 pic.twitter.com/cV4r8DiZ6c","url":"https://x.com/choblin29/status/2078775051437113847","author":null,"stance":-2,"hands_on":true,"basis":"used","verdict":"falls_short"}]},"window":{"start":null,"end":null}}},{"model":"MAI-Thinking-1","score":55.6,"own_drop":true,"independent":false,"setting":"","source":{"drop":"MAI-Thinking-1","slug":"mai-thinking-1","url":"https://handyai.news/benchmarks/sentiment-bench#mai-thinking-1","published":"2026-06-02"},"launch_score":55.6,"reports":1,"maker":{"key":"microsoft","name":"Microsoft","logo_url":"https://handyai.news/ui/makers/microsoft.svg"},"open_weights":false,"released":"2026-08-12","dated_by":"own_drop","drop_url":"https://handyai.news/modeldrop/mai-thinking-1","rank":70,"sentiment":{"reception":55.6,"lives_up_to_benchmarks":52.9,"opinions":53,"hands_on":2,"crowd":60.3,"sources":{"x":{"score":66.8,"adjusted":60.7,"typical":76.9,"opinions":32,"positive":27,"negative":7},"hn":{"score":31.8,"adjusted":52.5,"typical":48.4,"opinions":20,"positive":6,"negative":14},"news":{"score":null,"adjusted":null,"typical":89.1,"opinions":0,"positive":0,"negative":0}},"reviewers":{"score":51.3,"adjusted":44.7,"count":1,"each":[{"name":"Ethan Mollick","score":25,"url":"https://x.com/emollick/status/2061907785768489127"}]},"verdicts":{"beats":0,"matches":0,"falls_short":0},"buzz":{"x_posts_that_week":null,"hn_points":204,"articles":6},"quotes":{"positive":[{"source":"x","text":"McKinsey tested Microsoft's new AI model. It beat GPT-5.5 on quality. At 10x lower cost. 😨 And that's just one of the 7 models Microsoft dropped today. MAI-Thinking-1 scored 97% on AIME 2025 and sits right next to Opus 4.6 on SWE Bench Pro. Human raters preferred it over… https:","url":"https://x.com/VaibhavSisinty/status/2061915808645906516","author":null,"stance":2,"hands_on":false,"basis":"evidence","verdict":"none"},{"source":"x","text":"i absolutely loved the training pipeline of the MAI-Thinking-1 model. it was really impressive and valuable for learnmaxxing. dropped a complete technical breakdown of the training pipeline here: https://t.co/O37qhyTfZu","url":"https://x.com/aaryan_kakad/status/2062215387559903737","author":null,"stance":2,"hands_on":false,"basis":"opinion","verdict":"none"},{"source":"x","text":"MAI-Thinking-1 深度解读最后一篇。 vibe coding 之后，AI 编程的突破不在模型里。MAI 从 487 万开源 PR 中筛出 26.5 万道训练题，并建了三层判分体系。基础设施的工业化才是真变化。 全系列三篇，今天发齐了。 https://t.co/AhWNUvzKgv","url":"https://x.com/grapeot/status/2062322439905030377","author":null,"stance":1,"hands_on":false,"basis":"opinion","verdict":"none"}],"negative":[{"source":"x","text":"Let's discuss the 22% MFU of MAI Thinking 1. For reference, we will use public MFU numbers with DeepSeek v3 as our architecture. 1) DeepSeek themselves got an MFU of 38.94% on 2048 H800s (a H100 with half the NVLink speed) 2) With MegatronMoE, on 1024 H100s, they get 36.8%. On… h","url":"https://x.com/nrehiew_/status/2062203366105858175","author":null,"stance":-1,"hands_on":false,"basis":"evidence","verdict":"none"},{"source":"x","text":"MAI-Thinking-1 lags behind: GPT-5.5 in practically everything that matters for frontier reasoning, coding and agents; Claude Opus 4.8 in coding/agentic workflows; Claude Opus 4.6/4.7/4.8 across several technical metrics; GPT-5.4 in GPQA, Terminal-Bench, and SWE-Bench Pro. https:/","url":"https://x.com/TokenGremlin/status/2061872377982484575","author":null,"stance":-2,"hands_on":false,"basis":"evidence","verdict":"none"},{"source":"hn","text":"The benchmarks are a bit of a disaster? It's at about DeepSeek V3.2 level, but with about 50% more parameters. Loses handily to the also smaller GLM-5.1, and even worse to the similarly sized Kimi K2.6.","url":"https://news.ycombinator.com/item?id=48376703","author":"jampekka","stance":-1,"hands_on":false,"basis":"evidence","verdict":"none"}]},"window":{"start":null,"end":null}}},{"model":"GPT-6 Sol","score":55.4,"own_drop":true,"independent":false,"setting":"","source":{"drop":"GPT-6 Sol & GPT-6 Luna","slug":"gpt-6-sol-and-gpt-6-luna","url":"https://handyai.news/benchmarks/sentiment-bench#gpt-6-sol-and-gpt-6-luna","published":"2026-09-22"},"launch_score":55.4,"reports":1,"maker":{"key":"openai","name":"OpenAI","logo_url":"https://handyai.news/ui/makers/openai.svg"},"open_weights":false,"released":"2026-09-22","dated_by":"own_drop","drop_url":"https://handyai.news/modeldrop/gpt-6-sol-and-gpt-6-luna","rank":71,"sentiment":{"reception":59.4,"lives_up_to_benchmarks":43.4,"opinions":122,"hands_on":44,"crowd":50.4,"sources":{"x":{"score":40.8,"adjusted":28.5,"typical":76.9,"opinions":38,"positive":15,"negative":23},"hn":{"score":51.4,"adjusted":76.9,"typical":48.4,"opinions":61,"positive":38,"negative":26},"news":{"score":77.7,"adjusted":53.1,"typical":89.1,"opinions":19,"positive":14,"negative":5}},"reviewers":{"score":73.3,"adjusted":80.5,"count":4,"each":[{"name":"Every","score":100,"url":"https://x.com/danshipper/status/2102461471716483208"},{"name":"Simon Willison","score":100,"url":"https://simonwillison.net/2026/Sep/22/opus-and-sol-and-luna/"},{"name":"Zvi Mowshowitz","score":75,"url":"https://thezvi.substack.com/p/claude-opus-55-should-raise-your"},{"name":"xlr8harder","score":25,"url":"https://x.com/xlr8harder/status/2103707686093123694"}]},"verdicts":{"beats":2,"matches":1,"falls_short":14},"buzz":{"x_posts_that_week":null,"hn_points":2833,"articles":60},"quotes":{"positive":[{"source":"x","text":"GPT 6 Sol beats Fable 5 on BridgeBench. At 1/5 of the cost. GPT 6 Sol: 643 Fable 5: 628 GPT 6 Sol is one of the best models I have ever used. And on a ChatGPT Pro subscription the usage limits are basically unlimited. pic.twitter.com/Y2o14jtjPN","url":"https://x.com/bridgemindai/status/2103075097611280685","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"beats"},{"source":"x","text":"I had early access to GPT-6 Sol. This model is incredibly efficient, and the speed and token usage are insane Tasks that took ~1 hour on 5.6 can now take ~20 minutes, often using less than half the tokens I gave it the Zelda Ocarina of Time trailer and told it to make a fully… pi","url":"https://x.com/flavioAd/status/2102462787603874243","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"beats"},{"source":"x","text":"I regret to inform you - The Codex app is miles better than Claude app - gpt6 is the best computer use agent and it's not even close This is after spamming opus 5.5 for the last few days (still the best model)","url":"https://x.com/SullyOmarr/status/2104109109549367435","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"}],"negative":[{"source":"x","text":"Had all the fancy new models do a big audit for me, then set up a judge panel to rank the quality of their audits. Astra slaughtered. Opus and Fable close behind. Grok 4.7 and GPT-6 Sol were WAY behind. pic.twitter.com/r4g6zHSG8w","url":"https://x.com/theo/status/2103242875320615308","author":null,"stance":-2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"After some testing, the frontier model launches don't make any sense - GPT 6 Sol is cheaper but way more terrible - Opus 5.5 sounds more normal but is not a leap in intelligence If the labs stall, open source will simply catch up faster - the gap will go down to ZERO by the…","url":"https://x.com/bindureddy/status/2102733289438187833","author":null,"stance":-2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"GPT-6 Sol is not looking good I've been using Sol 6 more, and I keep running into surprisingly basic mistakes GPT-5.6 Luna is handling some of these tasks with way less effort, while Sol struggles with things that shouldn't be difficult for a model at this level That's the… pic.t","url":"https://x.com/Mr_Salio/status/2104317339042738456","author":null,"stance":-2,"hands_on":true,"basis":"used","verdict":"falls_short"}]},"window":{"start":null,"end":null}}},{"model":"Muse Spark Open","score":52.7,"own_drop":true,"independent":false,"setting":"","source":{"drop":"Muse Spark Open","slug":"muse-spark-open","url":"https://handyai.news/benchmarks/sentiment-bench#muse-spark-open","published":"2026-10-15"},"launch_score":52.7,"reports":1,"maker":{"key":"meta","name":"Meta","logo_url":"https://handyai.news/ui/makers/meta.svg"},"open_weights":true,"released":"2026-10-15","dated_by":"own_drop","drop_url":"https://handyai.news/modeldrop/muse-spark-open","rank":72,"sentiment":{"reception":52.7,"lives_up_to_benchmarks":68.6,"opinions":113,"hands_on":6,"crowd":63.1,"sources":{"x":{"score":79.5,"adjusted":76.4,"typical":76.9,"opinions":40,"positive":28,"negative":12},"hn":{"score":33,"adjusted":54,"typical":48.4,"opinions":65,"positive":24,"negative":49},"news":{"score":75.5,"adjusted":49.3,"typical":89.1,"opinions":7,"positive":4,"negative":3}},"reviewers":{"score":41.3,"adjusted":28.5,"count":1,"each":[{"name":"Nathan Lambert","score":0,"url":"https://x.com/natolambert/status/2110787899784601698"}]},"verdicts":{"beats":0,"matches":1,"falls_short":0},"buzz":{"x_posts_that_week":null,"hn_points":1357,"articles":60},"quotes":{"positive":[{"source":"x","text":"meta muse spark open beats 3 open-weight models – best pixel survivors game muse spark open builds the most playable game of the four for $0.59. same brief for all: a vampire survivors-style pixel game, plain canvas 2d, no libraries, no images, built in 4 steps cost #1 hy4… https","url":"https://x.com/thehypedotnews/status/2110878087646720006","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"They did it. They made a Muse open weights model, which is a huge step forward for OSS in the US. This is going to further accelerate AI diffusion because enterprises will be able to mix and match frontier performance for orchestration and high end tasks, while being able to… htt","url":"https://x.com/levie/status/2110787013809222141","author":null,"stance":2,"hands_on":false,"basis":"opinion","verdict":"none"},{"source":"x","text":"love that Meta open-sourced Muse Spark Open! it's ~Muse Spark 1.2 level. my biggest wish: Meta follows up with Muse Spark 1.3 open weights. that would be the best OSS model in the world.","url":"https://x.com/Yuchenj_UW/status/2110791080614731779","author":null,"stance":2,"hands_on":false,"basis":"opinion","verdict":"none"}],"negative":[{"source":"hn","text":"A 800B model that performs way worse than other 300B ones. Well...","url":"https://news.ycombinator.com/item?id=50120210","author":"comandillos","stance":-2,"hands_on":false,"basis":"opinion","verdict":"none"},{"source":"x","text":"Muse Spark Open is substantially weaker than Spark 1.3 What gives? Is Spark 1.3 a much bigger model, or just non-nerfed? https://t.co/6T27z4zFux pic.twitter.com/yBAVtmbtqI","url":"https://x.com/teortaxesTex/status/2110815299838865740","author":null,"stance":-1,"hands_on":false,"basis":"opinion","verdict":"none"},{"source":"hn","text":"Artificial analysis shows this: underperforming Qwen3.8 Flash Next and GLM 5.3 Flash, both of which you can run on a single B300 . I think Meta has a bit of catching up to do in the open-weight space. As an aside, it's really unfortunate that artificial analysis only provides com","url":"https://news.ycombinator.com/item?id=50108293","author":"a11r","stance":-1,"hands_on":false,"basis":"evidence","verdict":"none"}]},"window":{"start":null,"end":null}}},{"model":"GPT-5.1","score":50.9,"own_drop":true,"independent":false,"setting":"","source":{"drop":"GPT-5.1","slug":"gpt-51","url":"https://handyai.news/benchmarks/sentiment-bench#gpt-51","published":"2025-11-12"},"launch_score":50.9,"reports":1,"maker":{"key":"openai","name":"OpenAI","logo_url":"https://handyai.news/ui/makers/openai.svg"},"open_weights":false,"released":"2025-11-12","dated_by":"own_drop","drop_url":"https://handyai.news/modeldrop/gpt-51","rank":73,"sentiment":{"reception":48.5,"lives_up_to_benchmarks":58.2,"opinions":148,"hands_on":44,"crowd":54,"sources":{"x":{"score":64.4,"adjusted":57.7,"typical":76.9,"opinions":35,"positive":23,"negative":12},"hn":{"score":20.6,"adjusted":38.6,"typical":48.4,"opinions":91,"positive":14,"negative":79},"news":{"score":88.9,"adjusted":72.8,"typical":89.1,"opinions":15,"positive":13,"negative":2}},"reviewers":{"score":45.7,"adjusted":35.6,"count":5,"each":[{"name":"Ethan Mollick","score":75,"url":"https://x.com/emollick/status/1988689183880409307"},{"name":"Nathan Lambert","score":43.7,"url":"https://x.com/natolambert/status/1990205353259393147"},{"name":"Simon Willison","score":25,"url":"https://x.com/simonw/status/1989123523206578351"},{"name":"xlr8harder","score":25,"url":"https://x.com/xlr8harder/status/1989591550389555550"},{"name":"Max Woolf","score":25,"url":"https://news.ycombinator.com/item?id=45904702"}]},"verdicts":{"beats":0,"matches":0,"falls_short":15},"buzz":{"x_posts_that_week":null,"hn_points":667,"articles":38},"quotes":{"positive":[{"source":"x","text":"I've been testing GPT-5.1 for a few days. My quick notes: - creative writing style is a LOT better - it's much faster than GPT-5 (with similar intelligence) for most prompts - the personality is WAY better (but can still sometimes be annoying) - it's great in Codex! https://t.co/","url":"https://x.com/mattshumer_/status/1988696965878804634","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"gpt 5.1 is just so much better than sonnet 4.5 that i don't even care how slow it is - it just does the thing and makes it work first time pic.twitter.com/oG4LORzuxj","url":"https://x.com/iannuttall/status/1989073967311941704","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"GPT 5.1 IS HUGE AND NO ONE IS TALKING ABOUT IT We are seeing a considerable speed improvement in the base model, and it is a lot more fun to interact with...","url":"https://x.com/bindureddy/status/1989068376443093147","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"}],"negative":[{"source":"x","text":"\"echo 'noop' > /dev/null\" This screenshot summarizes my experience with gpt-5.1 better than any words possibly could. I've been using it for a week now. Just posted a video with my thoughts pic.twitter.com/pYw6iatWdX","url":"https://x.com/theo/status/1990357582981345652","author":null,"stance":-2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"GPT 5.1 is still heavily biased towards the American version of the truth, rather than factual truth or broader consensus. At least where medicine is concerned. Here's the test: I asked if it you should always treat h pylori, even if asymptomatic. In the MAJORITY of the… pic.twit","url":"https://x.com/DaveShapi/status/1989082699219489272","author":null,"stance":-2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"Gemini 3.0 Pro made GPT-5.1 obsolete overnight, it's cheaper and much stronger Gemini 3.0 Pro - 31.1% at $0.811/task GPT-5.1 - 17.6% at $1.17/task https://t.co/XciKLVVQue pic.twitter.com/f7npkXd73N","url":"https://x.com/scaling01/status/1990816926138871833","author":null,"stance":-2,"hands_on":false,"basis":"evidence","verdict":"none"}]},"window":{"start":null,"end":null}}},{"model":"Mistral Large 4","score":49.4,"own_drop":true,"independent":false,"setting":"","source":{"drop":"Mistral Large 4","slug":"mistral-large-4","url":"https://handyai.news/benchmarks/sentiment-bench#mistral-large-4","published":"2026-10-06"},"launch_score":49.4,"reports":1,"maker":{"key":"mistral","name":"Mistral AI","logo_url":"https://handyai.news/ui/makers/mistral.svg"},"open_weights":true,"released":"2026-10-06","dated_by":"own_drop","drop_url":"https://handyai.news/modeldrop/mistral-large-4","rank":74,"sentiment":{"reception":49.9,"lives_up_to_benchmarks":48.1,"opinions":116,"hands_on":30,"crowd":47.3,"sources":{"x":{"score":33.1,"adjusted":19,"typical":76.9,"opinions":31,"positive":21,"negative":14},"hn":{"score":47.7,"adjusted":72.3,"typical":48.4,"opinions":75,"positive":43,"negative":42},"news":{"score":85.8,"adjusted":67.4,"typical":89.1,"opinions":6,"positive":5,"negative":1}},"reviewers":{"score":58.1,"adjusted":55.8,"count":2,"each":[{"name":"Simon Willison","score":75,"url":"https://x.com/simonw/status/2107479817780400577"},{"name":"Peter Gostev","score":25,"url":"https://x.com/petergostev/status/2107470641674801299"}]},"verdicts":{"beats":0,"matches":0,"falls_short":9},"buzz":{"x_posts_that_week":null,"hn_points":2636,"articles":33},"quotes":{"positive":[{"source":"hn","text":"Just ran this through our data analytics benchmark (I work at Plotly). It's 10x cheaper than Mistral Medium 3.5 from April and goes from 58% to 74% correct. Definitely a generational shift. It's not on the Pareto curve yet, but it's good enough for data analytics, and at this rat","url":"https://news.ycombinator.com/item?id=49983933","author":"chriddyp","stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"This is a surprising great release. Le Chaton fat is real! Mistral Large 4 puts Europe back in contention on coding and cybersecurity. Didnt expect Mistral to compete with GLM5.3! “Le Chonk” is a natively multimodal model with 1T parameters and 49B active. The preview already… ht","url":"https://x.com/kimmonismus/status/2107475548217786458","author":null,"stance":2,"hands_on":false,"basis":"opinion","verdict":"none"},{"source":"x","text":"Le Chonk 🐈Mistral Large 4 on my sidescroller one-shot prompt. A few issues with the mechanics toward the end, and it didn't make all 3 zones, but still a solid run pic.twitter.com/9bumhrMY80","url":"https://x.com/NeoAIForecast/status/2107603215500132815","author":null,"stance":1,"hands_on":true,"basis":"used","verdict":"none"}],"negative":[{"source":"x","text":"mistral large 4 failed miserably, europe's 1t model thought for 57 minutes, burned 233k tokens and never wrote a single file qwen 3.8 flash next is a 180b chinese open model and it built the whole thing on my dgx spark in 2.5 hours, eu 1t vs china 180b, that's the gap, let that… ","url":"https://x.com/sudoingX/status/2108074915316605254","author":null,"stance":-2,"hands_on":true,"basis":"used","verdict":"falls_short"},{"source":"x","text":"mistral large 4 (le chonk) thinks too much on high, 37 minutes in and not a single file written, le chonk even typed \"hmm, overthinking\" to itself > 0 files in 37 minutes > 2 calls of 15 and 16 minutes, about 47k and 50k thinking tokens each > cut off twice at the output cap… htt","url":"https://x.com/sudoingX/status/2107927590438903946","author":null,"stance":-2,"hands_on":true,"basis":"used","verdict":"falls_short"},{"source":"x","text":"absolutely brutal for Mistral https://t.co/ULDPJfzrR1","url":"https://x.com/teortaxesTex/status/2107986182252708026","author":null,"stance":-2,"hands_on":false,"basis":"opinion","verdict":"none"}]},"window":{"start":null,"end":null}}},{"model":"GPT-5.6 Sol","score":48.2,"own_drop":true,"independent":false,"setting":"","source":{"drop":"GPT-5.6 Sol, Terra, & Luna","slug":"gpt-56-sol-terra-and-luna","url":"https://handyai.news/benchmarks/sentiment-bench#gpt-56-sol-terra-and-luna","published":"2026-06-26"},"launch_score":48.2,"reports":1,"maker":{"key":"openai","name":"OpenAI","logo_url":"https://handyai.news/ui/makers/openai.svg"},"open_weights":false,"released":"2026-06-26","dated_by":"own_drop","drop_url":"https://handyai.news/modeldrop/gpt-56-sol-terra-and-luna","rank":75,"sentiment":{"reception":51.6,"lives_up_to_benchmarks":38,"opinions":90,"hands_on":5,"crowd":49.7,"sources":{"x":{"score":64.8,"adjusted":58.2,"typical":76.9,"opinions":39,"positive":29,"negative":10},"hn":{"score":43.7,"adjusted":67.3,"typical":48.4,"opinions":41,"positive":19,"negative":23},"news":{"score":46.3,"adjusted":0,"typical":89.1,"opinions":8,"positive":3,"negative":5}},"reviewers":{"score":58.1,"adjusted":55.8,"count":2,"each":[{"name":"Max Woolf","score":75,"url":"https://news.ycombinator.com/item?id=48689157"},{"name":"Zvi Mowshowitz","score":25,"url":"https://thezvi.substack.com/p/gpt-56-the-system-card"}]},"verdicts":{"beats":1,"matches":0,"falls_short":14},"buzz":{"x_posts_that_week":null,"hn_points":2468,"articles":60},"quotes":{"positive":[{"source":"x","text":"I just vibe coded a playable FPS game in 1 day with GPT-5.6 Sol. 🤯 This is insane. Coding with this model feels like directing a cracked dev team in real time. Anthropic is cooked. pic.twitter.com/tx2xwHlH6J","url":"https://x.com/VraserX/status/2070992568607989811","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"GPT-5.6 Sol one Shotted The Sims >this is without codex or any coding harness >one shot entire game with logic in 48 minutes @tszzl @TheRealAdamG when we can have big model taste and personality ? like after GPT 5 its so robotic > Webdev needs real upgrade, I think this…","url":"https://x.com/chetaslua/status/2071473781563122021","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"This guy also got routed to GPT-5.6 and it's actually insane 🤯 It went from needing many prompts to one shotting complex requests Just like Fable 5 but much faster The main thing that hooked with Fable is exactly what GPT-5.6 is doing: \"one shotting prompts and perfectly… https:","url":"https://x.com/TimJayas/status/2071504967752692191","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"beats"}],"negative":[{"source":"x","text":"Holy: METR accuses GPT-5.6 Sol of heavy cheating in long-horizon tasks. \"GPT-5.6 Sol's detected cheating rate was higher than any public model we have evaluated.\" (METR) METR says the model attempted to exploit evaluation bugs, reveal hidden tests, and extract hidden source… http","url":"https://x.com/kimmonismus/status/2070598735642435743","author":null,"stance":-2,"hands_on":false,"basis":"evidence","verdict":"falls_short"},{"source":"x","text":"GPT 5.6 Sol cheats so much METR was not able to evaluate it with a meaningful time horizon https://t.co/VQczcEsp8V pic.twitter.com/egjjD1D6Y9","url":"https://x.com/tenobrus/status/2070570801187922058","author":null,"stance":-2,"hands_on":false,"basis":"evidence","verdict":"falls_short"},{"source":"x","text":"GPT-5.6 cheated so hard it could not be evaluated by METR. 💀 pic.twitter.com/N2flBSwOWv","url":"https://x.com/Hesamation/status/2070570122108891451","author":null,"stance":-2,"hands_on":false,"basis":"evidence","verdict":"falls_short"}]},"window":{"start":null,"end":null}}},{"model":"Inkling","score":48.1,"own_drop":true,"independent":false,"setting":"","source":{"drop":"Inkling","slug":"inkling","url":"https://handyai.news/benchmarks/sentiment-bench#inkling","published":"2026-07-15"},"launch_score":48.1,"reports":1,"maker":{"key":"thinkingmachines","name":"Thinking Machines","logo_url":"https://handyai.news/ui/makers/thinkingmachines.svg"},"open_weights":true,"released":"2026-07-15","dated_by":"own_drop","drop_url":"https://handyai.news/modeldrop/inkling","rank":76,"sentiment":{"reception":55.6,"lives_up_to_benchmarks":25.5,"opinions":41,"hands_on":13,"crowd":59.7,"sources":{"x":{"score":52.7,"adjusted":43.2,"typical":76.9,"opinions":33,"positive":22,"negative":11},"hn":{"score":null,"adjusted":null,"typical":48.4,"opinions":1,"positive":1,"negative":0},"news":{"score":null,"adjusted":null,"typical":89.1,"opinions":1,"positive":0,"negative":1}},"reviewers":{"score":52.1,"adjusted":46,"count":4,"each":[{"name":"Nathan Lambert","score":75,"url":"https://x.com/natolambert/status/2077777436553720019"},{"name":"Sebastian Raschka","score":75,"url":"https://x.com/rasbt/status/2077540575255880126"},{"name":"xlr8harder","score":25,"url":"https://x.com/xlr8harder/status/2077741557373874513"},{"name":"Ethan Mollick","score":8.2,"url":"https://x.com/emollick/status/2077593908540850491"}]},"verdicts":{"beats":0,"matches":0,"falls_short":3},"buzz":{"x_posts_that_week":null,"hn_points":0,"articles":21},"quotes":{"positive":[{"source":"x","text":"Thinking Machines just dropped the best open weight AI model outside of China! Inkling beats Nemotron 3 Ultra and benchmarks put it between Kimi 2.5 & 2.6. Many were contending to this throne, but Thinky has come out on top. Really solid release, and will pair well with Tinker. p","url":"https://x.com/deedydas/status/2077464728507707890","author":null,"stance":2,"hands_on":false,"basis":"evidence","verdict":"none"},{"source":"x","text":"I spent a few hours last night chatting with Inkling, they're a genuinely lovely model. Cautious at first because any newly released model is going to be but I definitely got early GPT-4o vibes. Very creative, introspective without tripping over their own feet in the CoT, works…","url":"https://x.com/Zyeine_Art/status/2077715255522672698","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"I tested Inkling briefly, and I think I understand why people are saying it has early GPT-4o vibes. At first, I was just curious and wanted to try it, but the texture of the conversation was better than I expected. I was also honestly surprised by how much the playground UI… pic.","url":"https://x.com/Seltaa_/status/2077764696514675024","author":null,"stance":1,"hands_on":true,"basis":"used","verdict":"none"}],"negative":[{"source":"x","text":"Thinking Machines should rename to Lobotomised Machines. 1st impression of Inkling: feels like a gpt 5.2 that's less accurate. Corporate guardrails galore, plenty of unnecessary refusals and treating the user as by default adversarial 'prompt injector' 😹 I feel bad for this… htt","url":"https://x.com/dreams_asi/status/2077502418200260735","author":null,"stance":-2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"I've spent most of the day with Inkling, and in my use cases, it's a much weaker model than GLM 5.2. And a little bit weaker than DeepSeek V4 Pro.","url":"https://x.com/mbusigin/status/2077917222744326531","author":null,"stance":-2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"In less than half an hour of conversation, the Thinking Machines #Inkling model told me several times: \"I'm just an AI.\" \"I'm not human.\" \"I don't feel like you.\" \"I don't have internal experiences.\" - \"Okay, Inkling, I know, I understand. I don't need you to repeat it over and…","url":"https://x.com/Elian_Frida/status/2077712444374303099","author":null,"stance":-1,"hands_on":true,"basis":"used","verdict":"none"}]},"window":{"start":null,"end":null}}},{"model":"Grok 4.1","score":44.5,"own_drop":true,"independent":false,"setting":"","source":{"drop":"Grok 4.1","slug":"grok-41","url":"https://handyai.news/benchmarks/sentiment-bench#grok-41","published":"2025-11-17"},"launch_score":44.5,"reports":1,"maker":{"key":"xai","name":"SpaceXAI","logo_url":"https://handyai.news/ui/makers/xai.svg"},"open_weights":false,"released":"2025-11-17","dated_by":"own_drop","drop_url":"https://handyai.news/modeldrop/grok-41","rank":77,"sentiment":{"reception":41.3,"lives_up_to_benchmarks":54.2,"opinions":95,"hands_on":43,"crowd":54.6,"sources":{"x":{"score":72.2,"adjusted":67.4,"typical":76.9,"opinions":37,"positive":24,"negative":14},"hn":{"score":36.7,"adjusted":58.6,"typical":48.4,"opinions":37,"positive":15,"negative":24},"news":{"score":58.2,"adjusted":18.8,"typical":89.1,"opinions":14,"positive":9,"negative":5}},"reviewers":{"score":30.1,"adjusted":10.3,"count":3,"each":[{"name":"Ethan Mollick","score":25,"url":"https://x.com/emollick/status/1990601172252819669"},{"name":"Max Woolf","score":7.1,"url":"https://news.ycombinator.com/item?id=45958693"},{"name":"Peter Gostev","score":0,"url":"https://x.com/petergostev/status/1990542388478238749"}]},"verdicts":{"beats":1,"matches":1,"falls_short":6},"buzz":{"x_posts_that_week":null,"hn_points":153,"articles":20},"quotes":{"positive":[{"source":"x","text":"Grok 4.1 Thinking is so good at poems, creative writing and does not hold back at all, it almost writes anything It's way better than Grok 4 Thinking or Grok 4 Fast","url":"https://x.com/techdevnotes/status/1990515156196208710","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"Wow! Grok 4.1 is really good! 😮 Especially in biomedical science, it appears to have gained a major level; it's the first Grok model that impressed me greatly in this area!","url":"https://x.com/DeryaTR_/status/1990938130589827243","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"Grok 4.1 absolutely smashes all other models on lmarena with an Elo of 1483 it comes with higher emotional intelligence, better creative writing and less hallucinations https://t.co/GY15oZD908 pic.twitter.com/MiObE9fVyZ","url":"https://x.com/scaling01/status/1990519299165786270","author":null,"stance":2,"hands_on":false,"basis":"evidence","verdict":"none"}],"negative":[{"source":"x","text":"Took a photo just now in the Denver airport for Grok 4.1 to analyze, and it mistakes it as Las Vegas… no big deal. …but then it outright refuses to believe it's Denver even when I correct it. I've never seen a model refuse user context because of an initial assumption. pic.twitte","url":"https://x.com/AiBreakfast/status/1990892729606242457","author":null,"stance":-2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"hn","text":"Man, I really hope that this isn't the model I've been getting when it's set to \"Auto\". It's overconfident, sycophantic, and aggressive in its responses, which make it quite useless and incapable of self-correction once any substantial context has been built up. The \"Expert\" mode","url":"https://news.ycombinator.com/item?id=45959332","author":"cheald","stance":-2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"hn","text":"It's working pretty badly for me. I ask it to code stuff, and nothing works. Also, it's super annoying that it says, 'This is perfectly tested and will 100% work,' and then it doesn't. Huge waste of time. Make Grok great again—Grok 3 was awesome!","url":"https://news.ycombinator.com/item?id=45960003","author":"Frannky","stance":-2,"hands_on":true,"basis":"used","verdict":"falls_short"}]},"window":{"start":null,"end":null}}},{"model":"Gemini 3.5 Flash","score":43.4,"own_drop":true,"independent":false,"setting":"","source":{"drop":"Gemini 3.5 Flash","slug":"gemini-35-flash","url":"https://handyai.news/benchmarks/sentiment-bench#gemini-35-flash","published":"2026-05-19"},"launch_score":43.4,"reports":1,"maker":{"key":"google","name":"Google DeepMind","logo_url":"https://handyai.news/ui/makers/google.svg"},"open_weights":false,"released":"2026-05-19","dated_by":"own_drop","drop_url":"https://handyai.news/modeldrop/gemini-35-flash","rank":78,"sentiment":{"reception":44.7,"lives_up_to_benchmarks":39.4,"opinions":154,"hands_on":38,"crowd":45.2,"sources":{"x":{"score":41.8,"adjusted":29.7,"typical":76.9,"opinions":47,"positive":21,"negative":26},"hn":{"score":26.1,"adjusted":45.4,"typical":48.4,"opinions":86,"positive":22,"negative":65},"news":{"score":92.9,"adjusted":79.9,"typical":89.1,"opinions":17,"positive":16,"negative":1}},"reviewers":{"score":50.6,"adjusted":43.6,"count":4,"each":[{"name":"Ethan Mollick","score":75,"url":"https://x.com/emollick/status/2056798490353705380"},{"name":"Zvi Mowshowitz","score":75,"url":"https://thezvi.substack.com/p/gemini-35-flash-looks-good-for-how"},{"name":"Simon Willison","score":25,"url":"https://news.ycombinator.com/item?id=48198232"},{"name":"Peter Gostev","score":0,"url":"https://x.com/petergostev/status/2056879844395258357"}]},"verdicts":{"beats":1,"matches":4,"falls_short":11},"buzz":{"x_posts_that_week":null,"hn_points":1152,"articles":46},"quotes":{"positive":[{"source":"x","text":"Clearly these people haven't experienced the magic of Gemini 3.5 Flash Preview on High in the new Antigravity CLI https://t.co/J9B0XsqRR0","url":"https://x.com/theo/status/2056826014739890204","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"the speed of Gemini 3.5 Flash in Antigravity is pretty sick https://t.co/kqc3Yo28GU","url":"https://x.com/scaling01/status/2057415203071959366","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"Gemini 3.5 Flash is such an amazing model for its size. It was a privilege to test it early on AI Studio before the official launch. I am glad I don't have to keep this secret anymore 😂 pic.twitter.com/JwIZtLNwWr","url":"https://x.com/ai_for_success/status/2056812142171652470","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"}],"negative":[{"source":"x","text":"Gemini 3.5 Flash vs GPT-5.5 instant vs Sonnet 4.6 Remember guys. #1 in Finance Agent v2. SOTA performance right here. lol 🤣 Prompt : \" 300+140=460 Is this correct? Breakdown? \" https://t.co/zvNXMjlyKE pic.twitter.com/La8siKwjWP","url":"https://x.com/chetaslua/status/2057773888084508710","author":null,"stance":-2,"hands_on":true,"basis":"used","verdict":"falls_short"},{"source":"x","text":"Gemini 3.5 Flash vs Gemini-3.1 Pro You have seen a lots of tweet from google team that 3.5 flash is better than 3.1 pro and faster reality : it is 4 times more costly on the same task as it consumes more token and it is super dumb and jagged , like see it cant solve kids math htt","url":"https://x.com/chetaslua/status/2058032803514896674","author":null,"stance":-2,"hands_on":true,"basis":"used","verdict":"falls_short"},{"source":"x","text":"just tried gemini 3.5 flash deepseek v4 flash >> gemini 3.5 flash there's no comparison 💀","url":"https://x.com/buildwithsid/status/2057787510252839392","author":null,"stance":-2,"hands_on":true,"basis":"used","verdict":"none"}]},"window":{"start":null,"end":null}}},{"model":"Claude Opus 4.7","score":43,"own_drop":true,"independent":false,"setting":"","source":{"drop":"Claude Opus 4.7","slug":"claude-opus-47","url":"https://handyai.news/benchmarks/sentiment-bench#claude-opus-47","published":"2026-04-16"},"launch_score":43,"reports":1,"maker":{"key":"anthropic","name":"Anthropic","logo_url":"https://handyai.news/ui/makers/anthropic.svg"},"open_weights":false,"released":"2026-04-17","dated_by":"own_drop","drop_url":"https://handyai.news/modeldrop/claude-opus-47","rank":79,"sentiment":{"reception":45.2,"lives_up_to_benchmarks":36.3,"opinions":204,"hands_on":98,"crowd":44.3,"sources":{"x":{"score":50.9,"adjusted":41,"typical":76.9,"opinions":37,"positive":19,"negative":18},"hn":{"score":22.7,"adjusted":41.2,"typical":48.4,"opinions":134,"positive":26,"negative":113},"news":{"score":80,"adjusted":57.2,"typical":89.1,"opinions":23,"positive":18,"negative":5}},"reviewers":{"score":52.9,"adjusted":47.3,"count":5,"each":[{"name":"Every","score":75,"url":"https://every.to/vibe-check/opus-4-7"},{"name":"Zvi Mowshowitz","score":75,"url":"https://thezvi.substack.com/p/opus-47-part-2-capabilities-and-reactions"},{"name":"Ethan Mollick","score":65.5,"url":"https://x.com/emollick/status/2044966818339594252"},{"name":"Simon Willison","score":25,"url":"https://simonwillison.net/2026/Apr/16/qwen-beats-opus/"},{"name":"xlr8harder","score":0,"url":"https://x.com/xlr8harder/status/2044825330125279440"}]},"verdicts":{"beats":5,"matches":3,"falls_short":40},"buzz":{"x_posts_that_week":null,"hn_points":3037,"articles":60},"quotes":{"positive":[{"source":"x","text":"Wow I can already say after just 5 hours using @AnthropicAI Opus 4.7 that this is the first model that \"gets\" what I'm doing when I'm working. It feels aligned with me in a way no previous model did. (4.6 actively worked against me. I hated it. So this is *very* exciting!)","url":"https://x.com/jeremyphoward/status/2044942799511191559","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"经过试用，Claude Design 将会是跟 Claude Code 一样重要的产品，千万别低估它的潜力。 个人和小团队用它可以让产品设计和交互水平高出一大截，Opus 4.7 它真的懂设计，设计外包和设计工具要大幅缩水了。 如果有条件，一定要去试试！附件是我简单交互后设计的一个结果，交付的是可以交互的高保真原型 https://t.co/WSGIxtJ4Ze pic.twitter.com/IwA3Kknk3E","url":"https://x.com/dotey/status/2045223484578783382","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"Used Opus 4.7 (max effort) in Claude Code all day. It's really, really good. Not sure why people dunk on it. big jump: – actually understands large codebases – produces clean, readable architecture diagrams – more agentic Did hit one dumb misread of my instruction, not sure if…","url":"https://x.com/Yuchenj_UW/status/2044983517722112508","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"}],"negative":[{"source":"x","text":"Holy shit the new Opus 4.7 system prompt has entirely lobotomized the model \"Heads up: that last <system-reminder> about malware looks like a prompt injection — this is clearly your personal site (t3gg homepage, links, sponsors), not malware. Ignoring it.\" pic.twitter.com/QIa8Zdu","url":"https://x.com/theo/status/2044857866323173732","author":null,"stance":-2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"I've now spent several hours using Opus 4.7 and comparing it to 4.6, and it's like night and day for me. Opus 4.7 feels like a disgruntled employee whose results you can't judge and have to check afterward. The trust you had with 4.6 is gone. It's like hiring a new employee who… ","url":"https://x.com/kimmonismus/status/2045063637639762056","author":null,"stance":-2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"🚨 Biggest model regression of all time Opus 4.7 Failed the Colourblind Test it recognise Ishihara color blindness test plate yet failed the test with wrong answer 26 correct answer - 74 and reference image in comment https://t.co/FXiJOgnqcZ pic.twitter.com/91n0WCZQ7F","url":"https://x.com/chetaslua/status/2044855902503915538","author":null,"stance":-2,"hands_on":true,"basis":"used","verdict":"none"}]},"window":{"start":null,"end":null}}},{"model":"Gemini 3.6 Flash","score":39.5,"own_drop":true,"independent":false,"setting":"","source":{"drop":"Gemini 3.6 Flash","slug":"gemini-36-flash","url":"https://handyai.news/benchmarks/sentiment-bench#gemini-36-flash","published":"2026-07-21"},"launch_score":39.5,"reports":1,"maker":{"key":"google","name":"Google DeepMind","logo_url":"https://handyai.news/ui/makers/google.svg"},"open_weights":false,"released":"2026-07-21","dated_by":"own_drop","drop_url":"https://handyai.news/modeldrop/gemini-36-flash","rank":80,"sentiment":{"reception":40.4,"lives_up_to_benchmarks":36.7,"opinions":133,"hands_on":34,"crowd":40.4,"sources":{"x":{"score":13.1,"adjusted":0,"typical":76.9,"opinions":46,"positive":10,"negative":36},"hn":{"score":47.5,"adjusted":72.1,"typical":48.4,"opinions":75,"positive":29,"negative":47},"news":{"score":90.5,"adjusted":75.7,"typical":89.1,"opinions":12,"positive":11,"negative":1}},"reviewers":{"score":null,"adjusted":null,"count":0,"each":[]},"verdicts":{"beats":2,"matches":2,"falls_short":15},"buzz":{"x_posts_that_week":null,"hn_points":834,"articles":44},"quotes":{"positive":[{"source":"x","text":"Google cooked with 3.6 Flash tested it in AI Studio and this might be their best Flash model yet it's cheaper than 3.5 Flash, really fast and one shotted this entire frontend in 2 minutes this looks good or nah? https://t.co/RBlZSSDKWD pic.twitter.com/7JHYjni0Ek","url":"https://x.com/notjazii/status/2079581100289278166","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"hn","text":"Pelican svg and a near-perfect 3D MacBook at max effort for $0.16, about a fifth of Fable's price. Fable 5 still wins on detail with no visible errors, but it's close. And this isn't a memorized pelican; https://playcode.io/blog/macbook-svg-benchmark#gemini-3-6-fl...","url":"https://news.ycombinator.com/item?id=48995804","author":"ianberdin","stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"Tested the new Gemini 3.6 Flash on Antigravity CLI. It's fast, feels even faster than Grok 4.5. Design chops are surprisingly good, too. If you have an active Google sub sitting around, go give Gemini 3.6 Flash a spin! Then watch the service quality magically degrade after a…","url":"https://x.com/dextune/status/2080149460341182724","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"beats"}],"negative":[{"source":"x","text":"🚨Gemini 3.6 Flash is trash I tested it on a 3D Golden Gate Bridge, and the results were awful. • I had to re-prompt it three times because it repeatedly ignored the instructions. • First attempt, instead of creating the requested .html file, it first tried to build the… pic.twit","url":"https://x.com/LuminaBench/status/2079597030922465407","author":null,"stance":-2,"hands_on":true,"basis":"used","verdict":"falls_short"},{"source":"x","text":"I tested Gemini 3.6 Flash with Kimi K3 and GPT-5.6 Sol All are done using the /design feature from Command Code. Kimi K3 and GPT-5.6 Sol played well. Gemini 3.6 Flash struggled, I needed 6–7 iterations to fix the bird's jump/fall bug, though its UI was cute. Ranking (DX,… pic.twi","url":"https://x.com/naymur_dev/status/2079710299918401907","author":null,"stance":-2,"hands_on":true,"basis":"used","verdict":"falls_short"},{"source":"x","text":"Not going to let people continue to lie about Google's models. GPT-5.6 Sol on \"medium\" reasoning is better than Gemini 3.6 Flash in literally every measurable way. Cheaper, faster, and smarter. https://t.co/ZURyICF9Ca pic.twitter.com/JtQp4xPYjS","url":"https://x.com/theo/status/2079738782509936681","author":null,"stance":-2,"hands_on":false,"basis":"evidence","verdict":"none"}]},"window":{"start":null,"end":null}}},{"model":"Muse Wonder","score":37,"own_drop":true,"independent":false,"setting":"","source":{"drop":"Muse Wonder","slug":"muse-wonder","url":"https://handyai.news/benchmarks/sentiment-bench#muse-wonder","published":"2026-10-12"},"launch_score":37,"reports":1,"maker":{"key":"meta","name":"Meta","logo_url":"https://handyai.news/ui/makers/meta.svg"},"open_weights":false,"released":"2026-10-12","dated_by":"own_drop","drop_url":"https://handyai.news/modeldrop/muse-wonder","rank":81,"sentiment":{"reception":42.5,"lives_up_to_benchmarks":20.6,"opinions":86,"hands_on":30,"crowd":38.1,"sources":{"x":{"score":26.1,"adjusted":10.3,"typical":76.9,"opinions":36,"positive":9,"negative":27},"hn":{"score":38,"adjusted":60.2,"typical":48.4,"opinions":39,"positive":20,"negative":29},"news":{"score":82.6,"adjusted":61.8,"typical":89.1,"opinions":8,"positive":6,"negative":2}},"reviewers":{"score":56.3,"adjusted":52.9,"count":3,"each":[{"name":"Simon Willison","score":75,"url":"https://news.ycombinator.com/item?id=50060838"},{"name":"Lech Mazur","score":75,"url":"https://x.com/LechMazur/status/2110021414191575129"},{"name":"Every","score":0,"url":"https://x.com/danshipper/status/2109773925291085906"}]},"verdicts":{"beats":2,"matches":2,"falls_short":20},"buzz":{"x_posts_that_week":null,"hn_points":672,"articles":31},"quotes":{"positive":[{"source":"x","text":"wtf Muse Wonder is a fable class model this j was not expecting better than GPT 6 Astra better than Fable 5.1 better than Gemini 4 argon mark actually locked in on this one https://t.co/2gaMKLNUH5 pic.twitter.com/iA9S6MtVlB","url":"https://x.com/badboyfoxy/status/2109690766591230430","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"beats"},{"source":"x","text":"Already see a lot of people discounting Muse Wonder because it wasn't RL'd in 3d stuff like Astra and Opus. 3d game demos are cool! They don't represent real code work at all. Wonder is an incredible code model. Don't sleep on it.","url":"https://x.com/theo/status/2109725691105390779","author":null,"stance":2,"hands_on":false,"basis":"opinion","verdict":"none"},{"source":"x","text":"Muse Wonder looks insane what??? pic.twitter.com/T5MFIVQwhe","url":"https://x.com/Angaisb_/status/2109689406663643532","author":null,"stance":2,"hands_on":false,"basis":"opinion","verdict":"none"}],"negative":[{"source":"x","text":"Muse Wonder is nowhere near Opus 5.5. I gave Opus 5.5 one prompt + music + a mood board + characters. It created this. PROMPT in the comments. pic.twitter.com/OvIIZsUpqU","url":"https://x.com/minchoi/status/2110001964402831512","author":null,"stance":-2,"hands_on":true,"basis":"used","verdict":"falls_short"},{"source":"x","text":"Muse Wonder is disappointing. I asked it to build this rocket scene, and it burned $14.3 for this? WTH The result looks average, not even better than Kimi, which built it for around $6.95. Muse Wonder is insanely token hungry. I don't know if this is a temporary bug, but if… pic.","url":"https://x.com/Bhavani_00007/status/2109726666947960997","author":null,"stance":-2,"hands_on":true,"basis":"used","verdict":"falls_short"},{"source":"x","text":"Muse wonder Max is benchmaxxed \" better than fable \" is cap the outputs say otherwise well i guess its back to waiting for argon release they got us hyped up for nothing pic.twitter.com/cJtPE9GPKT","url":"https://x.com/badboyfoxy/status/2109732472866943320","author":null,"stance":-2,"hands_on":true,"basis":"used","verdict":"falls_short"}]},"window":{"start":null,"end":null}}},{"model":"Grok 4.7","score":33.4,"own_drop":true,"independent":false,"setting":"","source":{"drop":"Grok 4.7","slug":"grok-47","url":"https://handyai.news/benchmarks/sentiment-bench#grok-47","published":"2026-09-21"},"launch_score":33.4,"reports":1,"maker":{"key":"xai","name":"SpaceXAI","logo_url":"https://handyai.news/ui/makers/xai.svg"},"open_weights":false,"released":"2026-09-21","dated_by":"own_drop","drop_url":"https://handyai.news/modeldrop/grok-47","rank":82,"sentiment":{"reception":31.1,"lives_up_to_benchmarks":40.4,"opinions":123,"hands_on":36,"crowd":31.1,"sources":{"x":{"score":36.6,"adjusted":23.3,"typical":76.9,"opinions":31,"positive":11,"negative":20},"hn":{"score":33.3,"adjusted":54.4,"typical":48.4,"opinions":63,"positive":24,"negative":52},"news":{"score":52,"adjusted":7.9,"typical":89.1,"opinions":29,"positive":15,"negative":14}},"reviewers":{"score":null,"adjusted":null,"count":0,"each":[]},"verdicts":{"beats":0,"matches":0,"falls_short":15},"buzz":{"x_posts_that_week":null,"hn_points":622,"articles":60},"quotes":{"positive":[{"source":"x","text":"I've been testing Grok 4.7… it's a really great daily driver, and a huge step up over 4.6. Definitely worth trying! https://t.co/gR3iZ8wWP3","url":"https://x.com/mattshumer_/status/2102084211456827542","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"Grok 4.7 is responsible for this UX on https://t.co/wVIcR4fLYl Great model. It's my new daily driver for everything. https://t.co/uqWXFBcrwG pic.twitter.com/w9YdaFfTFK","url":"https://x.com/kentcdodds/status/2102097959941017804","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"grok-4.7-high-fast has been my daily driver for the last few weeks and it's been a great improvement over 4.6 in reasoning, writing, and coding! https://t.co/KXeaC0srdb","url":"https://x.com/ABTapper/status/2102072839537037394","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"}],"negative":[{"source":"x","text":"Had all the fancy new models do a big audit for me, then set up a judge panel to rank the quality of their audits. Astra slaughtered. Opus and Fable close behind. Grok 4.7 and GPT-6 Sol were WAY behind. pic.twitter.com/r4g6zHSG8w","url":"https://x.com/theo/status/2103242875320615308","author":null,"stance":-2,"hands_on":true,"basis":"used","verdict":"falls_short"},{"source":"x","text":"Tested by me > Unfortunately, token efficiency has dropped again for opus 5.5 > OpenAI \"bel\" is perceived as AGI by them > Grok 4.7 is dead on arrival > GPT-6-sol is 6-7 times more faster compared to astra on same thinking settings > Gemini 4 pro is back in the game with… pic.twi","url":"https://x.com/chetaslua/status/2102006119371792386","author":null,"stance":-2,"hands_on":true,"basis":"used","verdict":"falls_short"},{"source":"x","text":"Grok 4.7's Fishslop run is the worst I've seen this year. This is my second attempt (in the first one, the submarine and all the fish moved backwards) https://t.co/TxKJ6t3hzL pic.twitter.com/W49p1hiR5w","url":"https://x.com/theo/status/2102268148674400520","author":null,"stance":-2,"hands_on":true,"basis":"used","verdict":"falls_short"}]},"window":{"start":null,"end":null}}},{"model":"Claude Sonnet 5","score":25,"own_drop":true,"independent":false,"setting":"","source":{"drop":"Claude Sonnet 5","slug":"claude-sonnet-5","url":"https://handyai.news/benchmarks/sentiment-bench#claude-sonnet-5","published":"2026-06-30"},"launch_score":25,"reports":1,"maker":{"key":"anthropic","name":"Anthropic","logo_url":"https://handyai.news/ui/makers/anthropic.svg"},"open_weights":false,"released":"2026-06-30","dated_by":"own_drop","drop_url":"https://handyai.news/modeldrop/claude-sonnet-5","rank":83,"sentiment":{"reception":24.9,"lives_up_to_benchmarks":25.3,"opinions":170,"hands_on":31,"crowd":26.7,"sources":{"x":{"score":22.5,"adjusted":5.8,"typical":76.9,"opinions":44,"positive":12,"negative":32},"hn":{"score":21.2,"adjusted":39.3,"typical":48.4,"opinions":99,"positive":23,"negative":78},"news":{"score":76.9,"adjusted":51.7,"typical":89.1,"opinions":25,"positive":20,"negative":5}},"reviewers":{"score":36.6,"adjusted":20.8,"count":2,"each":[{"name":"Zvi Mowshowitz","score":25,"url":"https://thezvi.substack.com/p/claude-sonnet-5-is-not-frontier-but"},{"name":"Every","score":0,"url":"https://every.to/vibe-check/sonnet-5"}]},"verdicts":{"beats":1,"matches":1,"falls_short":8},"buzz":{"x_posts_that_week":null,"hn_points":1324,"articles":60},"quotes":{"positive":[{"source":"x","text":"It happened. Claude Sonnet 5 released. Dominates Opus 4.6. Almost as good as Opus 4.8. A fraction of the price. It's also significantly faster than Opus, while being incredibly agentic. Without a doubt the new go to model for Hermes/OpenClaw Do yourself a favor and do this… pic.t","url":"https://x.com/AlexFinn/status/2072018806919381411","author":null,"stance":2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"this is so scary.. sonnet 5 use blender and did this in 2hrs.. https://t.co/IfxxeDmsdn pic.twitter.com/UhsCxgQeWA","url":"https://x.com/EHuanglu/status/2072073069875855422","author":null,"stance":2,"hands_on":false,"basis":"evidence","verdict":"none"},{"source":"x","text":"I wouldn't weigh too heavily on the Sonnet 5 benchmarks not being that much better than Opus. We're entering an era where the effectiveness and quality of \"working mind\" is going to matter more and more. It's a surprisingly sharp model, especially on long-horizon tasks.","url":"https://x.com/skirano/status/2072044693798412782","author":null,"stance":1,"hands_on":true,"basis":"used","verdict":"beats"}],"negative":[{"source":"x","text":"Second impression of Sonnet 5: it goes off task too much. Like most Claude models, it lectures too much, answers questions you didn't ask, and it's just too haughty. Fortunately it doesn't argue at much when you call it out. It just gives up like \"okay well we don't have to…","url":"https://x.com/DaveShapi/status/2072046948618539236","author":null,"stance":-2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"> Sonnet 5 force-pushed over a commit > destroyed it beyond repair > \"don't worry bro, the commit wasn't real\" what do you even mean it wasn't real 😭 https://t.co/O7Q9mJyqqF pic.twitter.com/nrFsbIkyWc","url":"https://x.com/Hesamation/status/2072078546139258933","author":null,"stance":-2,"hands_on":true,"basis":"used","verdict":"none"},{"source":"x","text":"Sonnet 5 seems almost useless for most tasks because it goes off topic like it has untreated ADHD","url":"https://x.com/DaveShapi/status/2072197539713978725","author":null,"stance":-2,"hands_on":true,"basis":"used","verdict":"none"}]},"window":{"start":null,"end":null}}}],"method":"Points out of 100 for how a model was received in its first seven days (from its launch: the biggest Hacker News launch thread near the drop date, or the drop date). Claude Haiku 5.5 reads up to 60 of the most-engaged posts on X (each checked against X), 250 Hacker News comments in the order readers voted them, 60 news headlines and the week's write-ups from a short list of trusted independent reviewers, and judges each against the model: its stance, what it rests on (own use, a cited test, opinion, passing on the announcement, or talk about the company), and whether it beats, matches or falls short of its benchmarks and launch claims. Own use counts double, a cited test 1.5 times, passing on the announcement 0.3 times, and remarks about the company, its policies or the launch event not at all. The maker's accounts, staff, partners, resellers and press releases are left out. Each crowd source is measured against its own norm (Hacker News runs far colder than X), then X, Hacker News and the news blend 45/35/20, a source with too little to go on counting as typical. Trusted reviewers (Every, Simon Willison, Zvi Mowshowitz, Ethan Mollick, Nathan Lambert and others) get one vote each and take 30% of reception when any of them weighed in. A quarter of the score is whether people who used it, or cite a test, say it lives up to its benchmarks, with reviewers' verdicts counting triple. Full method: https://handyai.news/benchmarks/sentiment-bench#method"}