---
title: The benchmark board
url: https://handyai.news/benchmarks
generated_at: 2026-11-04T16:39:17.777Z
benchmarks: 194
models: 195
---

# The benchmark board | handyai.news

> 2536 scores for 195 models on 194 benchmarks, from 105 Model Drops since 2024-12-05.

Scores come from the benchmark section of each handyai.news Model Drop, mostly the makers' launch tables. When drops disagree, the newest drop wins, since later launches re-run rivals on the current version of a benchmark; the model's own launch number is kept as launch_score. Settings (effort level, harness) vary between makers, so compare within a board with care. Models without their own drop are dated by the first drop that cites them.

## Best model overall: Claude Opus 5.5

Handy AI's weigh-in folds all 194 leaderboards into one ranking (points out of 100). Full ranking: https://handyai.news/api/benchmarks/weigh-in/all.md, image: https://handyai.news/api/benchmarks/weigh-in/all.png.

1. Claude Opus 5.5 (Anthropic): 84.8
2. Claude Sonnet 5.5 (Anthropic): 84.3
3. Muse Wonder Full (Meta): 83.4
4. Kimi K3.1 Pro (Moonshot AI): 83.3
5. Gemini 4 Carbon (Google DeepMind): 81.6

Best by topic:

- Cyber: MAI-Cyber-1-Flash (56.3 points, 10 boards)
- Long context: Gemini 1.5 Pro (50.3 points, 4 boards)
- Multimodal: GPT-6 Astra (72.2 points, 11 boards)
- Coding: Claude Opus 5.5 (80.5 points, 44 boards)
- Math: Grok 4 (68 points, 13 boards)
- Work: Kimi K3.1 Pro (80.7 points, 15 boards)
- Agents: Claude Opus 5.5 (81 points, 21 boards)
- Reasoning: Claude Opus 5.5 (77.3 points, 12 boards)
- Instructions: Nemotron 3 Ultra (53.6 points, 4 boards)
- Knowledge: o3 (62 points, 9 boards)

## Who holds first place

- Anthropic: 47 (Sentiment Bench, Artificial Analysis Intelligence Index, SWE-bench Verified, Humanity's Last Exam (no tools), AIME 2025, Terminal-Bench 2.0, GDPval-AA, Terminal-Bench 4.0, SWE-bench Pro, MMLU-Pro, Terminal-Bench, OSWorld 2, MMMLU, TAU-bench Retail, ProgramBench, MMMU-Pro, NL2Repo, OSWorld 2 (offline subset), τ²-bench Retail, JobBench, Terminal-Bench 3.0, BabyVision (with tools), FrontierCode v1.1, AA-Briefcase v1.1, APEX-Agents, Agents' Last Exam v1, HealthBench Professional, OSWorld, Toolathlon-Verified, Terminal-Bench 2.1 (standard harness), AA-Briefcase, CursorBench 4.0, TAU-bench Airline, Artificial Analysis Coding Agent Index, FrontierCode 2.0, Vals Index v2.2, Frontier Finance, Artificial Analysis Coding Agent Index v2.0, BenchCAD (with tools), τ²-Bench, Chartography, CursorBench 3.2, Meta Internal Coding Bench, MLE-Bench, BioMysteryBench hard, BioMysteryBench human-solved, SWE-bench Pro v2 Hard)
- OpenAI: 39 (GPQA Diamond, ARC-AGI-2, Aider Polyglot, τ²-bench Telecom, Terminal-Bench Science, Epoch Capabilities Index, SWE-bench Multilingual, Codeforces, MMLU, GPQA, ExploitBench, GDPval, Chartography (with tools), DeepSearchQA, CursorBench, FrontierMath, FrontierSWE v2, Professional Capture-the-Flag Challenges, SWE-Lancer IC Diamond, PaperBench-10, LiveBench Coding, Factual error rate, MMMU-Pro Vision, SEC-Bench Pro, SimpleQA hallucination rate, AIME 2024 4/4 reliability, ARC-AGI-3, Codeforces 4/4 reliability, CVE-Bench, Cyber Range, Cybersecurity Capture The Flag Challenges, FrontierMath Tier 4, GPQA Diamond 4/4 reliability, MLE-Bench-30, MultiChallenge, OpenAI-MRCR, SWE-Lancer Diamond, τ²-bench Airline, CritPt v2 Hard)
- Google DeepMind: 20 (MMMU, ARC-AGI-1, AutomationBench-AA, Vals Index, Video-MMMU, HLE-Verified, SimpleQA Verified, MRCR (1M), Artificial Analysis Intelligence Index v5.0, IFBench (loose), SWE-sweep, Vals Finance Agent v2, Agents' Last Exam v1.1, GDM-MRCR v2, MathArena Apex, USAMO 2025, Arena AI text score, MATH, Natural2Code, Terminal-Bench Science (standard harness))
- Moonshot AI: 13 (BrowseComp, Humanity's Last Exam (with tools), OSWorld-Verified, Harvey Legal Agent Benchmark, AA-LCR, Agents' Last Exam, Terminal-Bench 4.0 (standard harness), GDP.pdf, ZeroBench (with tools), PinchBench, Agents' Last Exam (CLI), Artificial Analysis Cyber Index, OfficeQA Pro v2)
- Meta: 12 (DeepSWE v1.1, MCP Atlas, CharXiv Reasoning, AA-Omniscience hallucination rate, SWE-Marathon, MemoryBench, Personal Superintelligence Index, ChartQA, HealthBench Hard, HLE-Diamond (no tools), MathVista, MRCR 512K-1M)
- SpaceXAI: 10 (AIME 2024, SimpleQA, LMArena text Elo, HMMT 2025, CyBench, MASK dishonesty rate, Tokens per second, EQ-Bench3, EEBench, Creative Writing v3)
- NVIDIA: 5 (IFBench, MATH-500, RULER @ 1M, HMMT Feb25, SWE-Bench (OpenHands))
- Alibaba: 5 (BFCL v3, GDPval rubrics, PaperBench, SuperGPQA, τ²-Bench (Weighted))
- Mistral AI: 5 (CyberGym-E2E, ArenaHard, DocVQA, HumanEval, Dense200)
- Nanbeige: 4 (EnterpriseOps-Gym, Claw-Eval pass^3, DeepPlanning, SkillsBench)
- DeepSeek: 3 (LiveCodeBench, Apex Bench v1.1, Arena Hard)
- Liquid AI: 3 (IFEval, Tau3-Bench, Multi-IF)
- MiniMax: 2 (AutomationBench, BankerToolBench)
- Z.ai: 2 (AIME 2026, Finance Agent v2)
- Xiaomi: 1 (Terminal-Bench 2.1)
- Microsoft: 1 (CyberGym)
- Ant Group: 1 (SciCode)
- Mirendil: 1 (SWE-Atlas-QnA)
- Other: 1 (CyberGym Level 1 leaderboard)

## Biggest generational jumps

- Gemma 3 27B to Gemma 4 on τ²-bench Retail: 6.6% to 86.4% (+79.8 points)
- GPT-4.1 to GPT-5 on τ²-bench Telecom: 34% to 96.7% (+62.7 points)
- Claude Sonnet 5 to Claude Sonnet 5.5 on Terminal-Bench 4.0: 10.3% to 70.6% (+60.3 points)
- Claude Haiku 4.5 to Claude Haiku 5.5 on OSWorld 2 (offline subset): 15.7% to 72.4% (+56.7 points)
- Gemini 2.0 Flash to Gemini 2.5 Flash on AIME 2025: 27.5% to 78% (+50.5 points)
- GPT-5.6 Terra to GPT-6.1 Terra on AA-Omniscience hallucination rate: 88% to 38% (+50 points)
- Claude Sonnet 4 to Claude Sonnet 4.5 on τ²-bench Telecom: 49.6% to 98% (+48.4 points)
- Claude 3.5 Sonnet to Claude 3.7 Sonnet on AIME 2024: 16% to 61.3% (+45.3 points)

## Sentiment Bench

How people received the model in its first week on X, Hacker News, in the news and among trusted reviewers (Handy AI). 83 models from 79 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Claude Opus 4.5 | Anthropic | 91.5 |  | [Claude Opus 4.5](https://handyai.news/benchmarks/sentiment-bench#claude-opus-45) (own drop) |
| 2 | GLM 5.2 | Z.ai | 88 |  | [GLM 5.2](https://handyai.news/benchmarks/sentiment-bench#glm-52) (own drop) |
| 3 | Grok 4.6 | SpaceXAI | 87 |  | [Grok 4.6](https://handyai.news/benchmarks/sentiment-bench#grok-46) (own drop) |
| 4 | Gemini 4 Carbon | Google DeepMind | 85.8 |  | [Gemini 4 Carbon](https://handyai.news/benchmarks/sentiment-bench#gemini-4-carbon) (own drop) |
| 5 | GPT-5.3-Codex | OpenAI | 85.6 |  | [GPT-5.3-Codex](https://handyai.news/benchmarks/sentiment-bench#gpt-53-codex) (own drop) |

78 more: https://handyai.news/api/benchmarks/sentiment-bench

## Artificial Analysis Intelligence Index

138 models from 75 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Claude Opus 5.5 | Anthropic | 58 |  | [Kimi K3.1 Pro](https://handyai.news/modeldrop/kimi-k31-pro) |
| 2 | Claude Sonnet 5.5 | Anthropic | 56 | max | [GPT-6.1 Terra](https://handyai.news/modeldrop/gpt-61-terra) |
| 3 | Muse Wonder Full | Meta | 55.8 | max | [Muse Wonder](https://handyai.news/modeldrop/muse-wonder) |
| 4 | Kimi K3.1 Pro | Moonshot AI | 55 |  | [Kimi K3.1 Pro](https://handyai.news/modeldrop/kimi-k31-pro) (own drop) |
| 5 | GPT-6 Astra | OpenAI | 53 |  | [Kimi K3.1 Pro](https://handyai.news/modeldrop/kimi-k31-pro) |

133 more: https://handyai.news/api/benchmarks/artificial-analysis-intelligence-index

## GPQA Diamond

107 models from 59 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | GPT-6 Astra | OpenAI | 96.1% | max | [Kimi K3.1 Pro](https://handyai.news/modeldrop/kimi-k31-pro) |
| 2 | Claude Mythos Preview | Anthropic | 94.5% | max | [Claude Mythos Preview](https://handyai.news/modeldrop/claude-mythos-preview) (own drop) |
| 3 | Gemini 3.1 Pro | Google DeepMind | 94.3% | high | [Muse Spark](https://handyai.news/modeldrop/muse-spark) |
| 4 | Kimi K3.1 Pro | Moonshot AI | 94.2% | max | [Kimi K3.1 Pro](https://handyai.news/modeldrop/kimi-k31-pro) (own drop) |
| 5 | GPT-5.6 Sol | OpenAI | 94.1% | max | [Qwen3.8-Max](https://handyai.news/modeldrop/qwen38-max) |

102 more: https://handyai.news/api/benchmarks/gpqa-diamond

## SWE-bench Verified

75 models from 48 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Claude Mythos Preview | Anthropic | 93.9% | max | [Claude Mythos Preview](https://handyai.news/modeldrop/claude-mythos-preview) (own drop) |
| 2 | GPT-6 Astra | OpenAI | 86% |  | [Mirendil, "Introducing Fern"](https://mirendil.com/news/introducing-fern/) |
| 3 | Claude Opus 4.5 | Anthropic | 80.9% | high | [Qwen3.5](https://handyai.news/modeldrop/qwen35) |
| 4 | Claude Opus 4.6 | Anthropic | 80.8% | max | [MAI-Thinking-1](https://handyai.news/modeldrop/mai-thinking-1) |
| 5 | Gemini 3.1 Pro | Google DeepMind | 80.6% | high | [Muse Spark](https://handyai.news/modeldrop/muse-spark) |

70 more: https://handyai.news/api/benchmarks/swe-bench-verified

## Humanity's Last Exam (no tools)

Expert-level questions, no tools. 77 models from 40 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Claude Opus 5.5 | Anthropic | 64.4% |  | [Claude Opus 5.5](https://handyai.news/modeldrop/claude-opus-55) (own drop) |
| 2 | Mythos 5 | Anthropic | 59% |  | [Fable 5 / Mythos 5](https://handyai.news/modeldrop/fable-5-mythos-5) (own drop) |
| 3 | Fable 5 | Anthropic | 57.8% |  | [Fable 5.1](https://handyai.news/modeldrop/fable-51) |
| 4 | Fable 5.1 | Anthropic | 57.6% | max | [Kimi K3.1 Pro](https://handyai.news/modeldrop/kimi-k31-pro) |
| 5 | Claude Sonnet 5.5 | Anthropic | 56.9% |  | [Claude Haiku 5.5](https://handyai.news/modeldrop/claude-haiku-55) |

72 more: https://handyai.news/api/benchmarks/humanitys-last-exam-no-tools

## LiveCodeBench

61 models from 24 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | DeepSeek V4 | DeepSeek | 93.5% |  | [MAI-Thinking-1](https://handyai.news/modeldrop/mai-thinking-1) |
| 2 | Gemini 3 Pro | Google DeepMind | 90.7% |  | [Qwen3.5](https://handyai.news/modeldrop/qwen35) |
| 3 | Kimi K2.6 | Moonshot AI | 89.6% |  | [MAI-Thinking-1](https://handyai.news/modeldrop/mai-thinking-1) |
| 4 | Nemotron 3 Ultra | NVIDIA | 89% |  | [Nemotron 3 Ultra](https://handyai.news/modeldrop/nemotron-3-ultra) (own drop) |
| 5 | GPT-5.2 | OpenAI | 87.7% |  | [Qwen3.5](https://handyai.news/modeldrop/qwen35) |

56 more: https://handyai.news/api/benchmarks/livecodebench

## AIME 2025

46 models from 24 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Claude Opus 4.6 | Anthropic | 99.8% |  | [MAI-Thinking-1](https://handyai.news/modeldrop/mai-thinking-1) |
| 2 | o4-mini | OpenAI | 99.5% |  | [gpt-oss-120b & gpt-oss-20b](https://handyai.news/modeldrop/gpt-oss) |
| 3 | gpt-oss-120b | OpenAI | 97.9% |  | [gpt-oss-120b & gpt-oss-20b](https://handyai.news/modeldrop/gpt-oss) (own drop) |
| 4 | MAI-Thinking-1 | Microsoft | 97% |  | [MAI-Thinking-1](https://handyai.news/modeldrop/mai-thinking-1) (own drop) |
| 5 | Claude Sonnet 4.6 | Anthropic | 95.6% | max | [MAI-Thinking-1](https://handyai.news/modeldrop/mai-thinking-1) |

41 more: https://handyai.news/api/benchmarks/aime-2025

## DeepSWE v1.1

Fixing issues in real repositories. 45 models from 24 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Muse Wonder Full | Meta | 79% | max | [Muse Wonder](https://handyai.news/modeldrop/muse-wonder) |
| 2 | Muse Wonder | Meta | 78.8% | max | [Muse Wonder](https://handyai.news/modeldrop/muse-wonder) (own drop) |
| 3 | Gemini 4 Argon | Google DeepMind | 77.9% | max | [Gemini 4 Argon](https://handyai.news/modeldrop/gemini-4-argon) (own drop) |
| 4 | DeepSeek V4.1 Flash 1017 | DeepSeek | 77% | max | [Kimi K3.1 Pro](https://handyai.news/modeldrop/kimi-k31-pro) |
| 5 | Kimi K3.1 Pro | Moonshot AI | 75.5% | max | [Kimi K3.1 Pro](https://handyai.news/modeldrop/kimi-k31-pro) (own drop) |

40 more: https://handyai.news/api/benchmarks/deepswe-v1-1

## Terminal-Bench 2.0

Hard tasks in a real terminal (2025 edition). 33 models from 22 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Claude Mythos Preview | Anthropic | 82% | max | [Claude Mythos Preview](https://handyai.news/modeldrop/claude-mythos-preview) (own drop) |
| 2 | GPT-5.3-Codex | OpenAI | 77.3% |  | [GPT-5.4](https://handyai.news/modeldrop/gpt-54) |
| 3 | GPT-5.4 | OpenAI | 75.1% | xhigh | [MAI-Thinking-1](https://handyai.news/modeldrop/mai-thinking-1) |
| 4 | Gemini 3.1 Pro | Google DeepMind | 68.5% | high | [Muse Spark](https://handyai.news/modeldrop/muse-spark) |
| 5 | Claude Opus 4.6 | Anthropic | 65.4% | max | [Muse Spark](https://handyai.news/modeldrop/muse-spark) |

28 more: https://handyai.news/api/benchmarks/terminal-bench-2-0

## BrowseComp

49 models from 21 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Kimi K3.1 Pro | Moonshot AI | 92.1% | max | [Kimi K3.1 Pro](https://handyai.news/modeldrop/kimi-k31-pro) (own drop) |
| 2 | LongCat-2.5 | Meituan | 91.9% |  | [Meituan LongCat-2.5 launch post](https://x.com/Meituan_LongCat/status/2117264311714537633) (own drop) |
| 3 | GPT-6 Astra | OpenAI | 91.5% | max | [Kimi K3.1 Pro](https://handyai.news/modeldrop/kimi-k31-pro) |
| 4 | Claude Opus 5.5 | Anthropic | 90.8% | max | [Kimi K3.1 Pro](https://handyai.news/modeldrop/kimi-k31-pro) |
| 5 | GPT-5.6 Sol | OpenAI | 90.4% | max | [GPT-5.6 Sol, Terra, & Luna](https://handyai.news/modeldrop/gpt-56-sol-terra-and-luna) (own drop) |

44 more: https://handyai.news/api/benchmarks/browsecomp

## Terminal-Bench 2.1

47 models from 20 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | MiMo-V2.6-Pro | Xiaomi | 89.9% |  | [MiMo-V2.6-Pro](https://handyai.news/modeldrop/mimo-v26-pro) (own drop) |
| 2 | Gemini 3.8 Flash | Google DeepMind | 89.4% | max | [Gemini 3.8 Flash](https://handyai.news/modeldrop/gemini-38-flash) (own drop) |
| 3 | Claude Opus 5 | Anthropic | 89.1% |  | [MiMo-V2.6-Pro](https://handyai.news/modeldrop/mimo-v26-pro) |
| 4 | GPT-5.6 Sol | OpenAI | 88.8% | max | [MiMo-V2.6-Pro](https://handyai.news/modeldrop/mimo-v26-pro) |
| 5 | Muse Spark 1.3 | Meta | 88.8% | max | [Muse Spark 1.3](https://handyai.news/modeldrop/muse-spark-13) (own drop) |

42 more: https://handyai.news/api/benchmarks/terminal-bench-2-1

## ARC-AGI-2

34 models from 20 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | GPT-5.6 Sol | OpenAI | 92.5% |  | [Grok 4.6](https://handyai.news/modeldrop/grok-46) |
| 2 | Claude Opus 5 | Anthropic | 90.4% |  | [Grok 4.6](https://handyai.news/modeldrop/grok-46) |
| 3 | Fable 5 | Anthropic | 89.2% |  | [Grok 4.6](https://handyai.news/modeldrop/grok-46) |
| 4 | Gemini 3.8 Flash | Google DeepMind | 89.2% |  | [Gemini 3.8 Flash](https://handyai.news/modeldrop/gemini-38-flash) (own drop) |
| 5 | Gemini 3.7 Flash | Google DeepMind | 84.6% |  | [Gemini 3.7 Flash](https://handyai.news/modeldrop/gemini-37-flash) (own drop) |

29 more: https://handyai.news/api/benchmarks/arc-agi-2

## GDPval-AA

Professional work products, graded head to head (Elo). 39 models from 19 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Claude Opus 5.5 | Anthropic | 1866 Elo | max | [Kimi K3.1 Pro](https://handyai.news/modeldrop/kimi-k31-pro) |
| 2 | Muse Wonder Full | Meta | 1857 Elo | max | [Muse Wonder](https://handyai.news/modeldrop/muse-wonder) |
| 3 | Kimi K3.1 Pro | Moonshot AI | 1846 Elo | max | [Kimi K3.1 Pro](https://handyai.news/modeldrop/kimi-k31-pro) (own drop) |
| 4 | Claude Sonnet 5.5 | Anthropic | 1838 Elo |  | [Muse Wonder](https://handyai.news/modeldrop/muse-wonder) |
| 5 | Muse Wonder | Meta | 1770 Elo | max | [Muse Wonder](https://handyai.news/modeldrop/muse-wonder) (own drop) |

34 more: https://handyai.news/api/benchmarks/gdpval-aa

## Terminal-Bench 4.0

Multi-hour work in a real terminal. 37 models from 19 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Claude Sonnet 5.5 | Anthropic | 70.6% | max | [GPT-6.1 Terra](https://handyai.news/modeldrop/gpt-61-terra) |
| 2 | Claude Opus 5.5 | Anthropic | 66.4% | max | [Kimi K3.1 Pro](https://handyai.news/modeldrop/kimi-k31-pro) |
| 3 | Kimi K3.1 Pro | Moonshot AI | 65.7% | max | [Kimi K3.1 Pro](https://handyai.news/modeldrop/kimi-k31-pro) (own drop) |
| 4 | Mythos 5.1 | Anthropic | 60.9% |  | [GPT-6 Astra](https://handyai.news/modeldrop/gpt-6-astra) |
| 5 | GPT-6 Astra | OpenAI | 58.2% | max | [GPT-6.1 Terra](https://handyai.news/modeldrop/gpt-61-terra) |

32 more: https://handyai.news/api/benchmarks/terminal-bench-4-0

## MMMU

32 models from 17 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Gemini 3 Pro | Google DeepMind | 87.2% |  | [Qwen3.5](https://handyai.news/modeldrop/qwen35) |
| 2 | GPT-5.2 | OpenAI | 86.7% |  | [Qwen3.5](https://handyai.news/modeldrop/qwen35) |
| 3 | GPT-5.1 | OpenAI | 85.4% | high | [Claude Opus 4.5](https://handyai.news/modeldrop/claude-opus-45) |
| 4 | Qwen3.5 | Alibaba | 85% |  | [Qwen3.5](https://handyai.news/modeldrop/qwen35) (own drop) |
| 5 | Kimi K2.5 | Moonshot AI | 84.3% |  | [Qwen3.5](https://handyai.news/modeldrop/qwen35) |

27 more: https://handyai.news/api/benchmarks/mmmu

## ARC-AGI-1

28 models from 17 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Gemini 3.1 Pro | Google DeepMind | 98% |  | [Gemini 3.1 Pro](https://handyai.news/modeldrop/gemini-31-pro) (own drop) |
| 2 | GPT-5.6 Terra | OpenAI | 96.5% | max | [Gemini 3.7 Flash](https://handyai.news/modeldrop/gemini-37-flash) |
| 3 | Gemini 3.7 Flash | Google DeepMind | 95.5% |  | [Gemini 3.7 Flash](https://handyai.news/modeldrop/gemini-37-flash) (own drop) |
| 4 | Claude Opus 4.6 | Anthropic | 94% | high | [Gemini 3.1 Pro](https://handyai.news/modeldrop/gemini-31-pro) |
| 5 | Gemini 3.6 Flash | Google DeepMind | 91.2% | high | [Gemini 3.7 Flash](https://handyai.news/modeldrop/gemini-37-flash) |

23 more: https://handyai.news/api/benchmarks/arc-agi-1

## Aider Polyglot

27 models from 17 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | GPT-5 | OpenAI | 88% | high | [GPT-5](https://handyai.news/modeldrop/gpt-5) (own drop) |
| 2 | o3-pro | OpenAI | 84.9% |  | [o3-pro](https://handyai.news/modeldrop/o3-pro) (own drop) |
| 3 | o3 | OpenAI | 79.6% |  | [GPT-5](https://handyai.news/modeldrop/gpt-5) |
| 4 | Gemini 2.5 Pro | Google DeepMind | 76.9% |  | [DeepSeek R1 0528](https://handyai.news/modeldrop/deepseek-r1-0528) |
| 5 | DeepSeek V3.1 | DeepSeek | 76.3% | thinking | [DeepSeek V3.1](https://handyai.news/modeldrop/deepseek-v31) (own drop) |

22 more: https://handyai.news/api/benchmarks/aider-polyglot

## AutomationBench

Real business automations (Zapier). 31 models from 16 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | MiniMax M3.1 Flash | MiniMax | 62.2% |  | [MiniMax M3.1 Flash](https://handyai.news/modeldrop/minimax-m31-flash) (own drop) |
| 2 | DeepSeek V4.1 Flash 1017 | DeepSeek | 60.5% |  | [MiniMax M3.1 Flash](https://handyai.news/modeldrop/minimax-m31-flash) |
| 3 | Mistral Large 4 | Mistral AI | 59.9% |  | [Mistral Large 4](https://handyai.news/modeldrop/mistral-large-4) (own drop) |
| 4 | DeepSeek V4 Pro 0813 | DeepSeek | 56.7% |  | [Mistral Large 4](https://handyai.news/modeldrop/mistral-large-4) |
| 5 | Hy4 | Tencent | 56.2% |  | [Tencent Hunyuan Hy4 launch post](https://x.com/TencentHunyuan/status/2117870250838630539) (own drop) |

26 more: https://handyai.news/api/benchmarks/automationbench

## SWE-bench Pro

36 models from 15 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Fable 5 | Anthropic | 80% | max | [Qwen3.8](https://handyai.news/modeldrop/qwen38) |
| 2 | Claude Mythos Preview | Anthropic | 77.8% | max | [Claude Mythos Preview](https://handyai.news/modeldrop/claude-mythos-preview) (own drop) |
| 3 | Claude Opus 4.8 | Anthropic | 69.2% | max | [Qwen3.8](https://handyai.news/modeldrop/qwen38) |
| 4 | Qwen3.8-Max | Alibaba | 67.7% |  | [Qwen3.8-Max](https://handyai.news/modeldrop/qwen38-max) (own drop) |
| 5 | Grok 4.5 | SpaceXAI | 64.7% | max | [Gemini 3.6 Flash](https://handyai.news/modeldrop/gemini-36-flash) |

31 more: https://handyai.news/api/benchmarks/swe-bench-pro

## Humanity's Last Exam (with tools)

Expert-level questions, with search and code. 40 models from 14 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Kimi K3.1 Pro | Moonshot AI | 70% | max | [Kimi K3.1 Pro](https://handyai.news/modeldrop/kimi-k31-pro) (own drop) |
| 2 | Claude Opus 5.5 | Anthropic | 67.7% |  | [Claude Opus 5.5](https://handyai.news/modeldrop/claude-opus-55) (own drop) |
| 3 | Fable 5.1 | Anthropic | 65.6% | max | [Kimi K3.1 Pro](https://handyai.news/modeldrop/kimi-k31-pro) |
| 4 | Claude Mythos Preview | Anthropic | 64.7% |  | [Fable 5 / Mythos 5](https://handyai.news/modeldrop/fable-5-mythos-5) |
| 5 | Mythos 5 | Anthropic | 64.5% |  | [Fable 5 / Mythos 5](https://handyai.news/modeldrop/fable-5-mythos-5) (own drop) |

35 more: https://handyai.news/api/benchmarks/humanitys-last-exam-with-tools

## AIME 2024

23 models from 12 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Grok 4 | SpaceXAI | 94.3% |  | [GLM-4.5](https://handyai.news/modeldrop/glm-45) |
| 2 | o3-pro | OpenAI | 93% |  | [o3-pro](https://handyai.news/modeldrop/o3-pro) (own drop) |
| 3 | GLM-4.5 | Z.ai | 91% |  | [GLM-4.5](https://handyai.news/modeldrop/glm-45) (own drop) |
| 4 | o3 | OpenAI | 90.3% |  | [GLM-4.5](https://handyai.news/modeldrop/glm-45) |
| 5 | DeepSeek R1 0528 | DeepSeek | 89.3% |  | [GLM-4.5](https://handyai.news/modeldrop/glm-45) |

18 more: https://handyai.news/api/benchmarks/aime-2024

## τ²-bench Telecom

26 models from 11 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | GPT-5.2 | OpenAI | 98.7% | xhigh | [GPT-5.2](https://handyai.news/modeldrop/gpt-52) (own drop) |
| 2 | GLM-5.3-Flash | Z.ai | 98.5% |  | [Nemotron 3.5 Super VL](https://handyai.news/modeldrop/nemotron-35-super-vl) |
| 3 | Claude Sonnet 4.5 | Anthropic | 98% |  | [Claude Haiku 4.5](https://handyai.news/modeldrop/claude-haiku-45) |
| 4 | GPT-5 | OpenAI | 96.7% | high | [GPT-5.1](https://handyai.news/modeldrop/gpt-51) |
| 5 | GPT-5.1 | OpenAI | 95.6% | high | [GPT-5.1](https://handyai.news/modeldrop/gpt-51) (own drop) |

21 more: https://handyai.news/api/benchmarks/bench-telecom

## SimpleQA

23 models from 11 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Grok 4 Fast | SpaceXAI | 95% |  | [Grok 4 Fast](https://handyai.news/modeldrop/grok-4-fast) (own drop) |
| 2 | Grok 4 | SpaceXAI | 94% |  | [Grok 4 Fast](https://handyai.news/modeldrop/grok-4-fast) |
| 3 | Grok 3 | SpaceXAI | 82% |  | [Grok 4 Fast](https://handyai.news/modeldrop/grok-4-fast) |
| 4 | GPT-4.5 | OpenAI | 62.5% |  | [Gemini 2.5 Pro](https://handyai.news/modeldrop/gemini-25-pro) |
| 5 | Gemini 2.5 Pro | Google DeepMind | 52.9% |  | [Gemini 2.5 Pro](https://handyai.news/modeldrop/gemini-25-pro) (own drop) |

18 more: https://handyai.news/api/benchmarks/simpleqa

## OSWorld-Verified

Using a computer: apps, files and the web (2025 edition). 24 models from 10 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Kimi K3 | Moonshot AI | 84.8% |  | [MiMo-V2.6-Pro](https://handyai.news/modeldrop/mimo-v26-pro) |
| 2 | Qwen3.8 | Alibaba | 84.3% | max | [Qwen3.8](https://handyai.news/modeldrop/qwen38) (own drop) |
| 3 | Claude Opus 5 | Anthropic | 83.4% |  | [MiMo-V2.6-Pro](https://handyai.news/modeldrop/mimo-v26-pro) |
| 4 | Gemini 3.6 Flash | Google DeepMind | 83% | max | [Gemini 3.6 Flash](https://handyai.news/modeldrop/gemini-36-flash) (own drop) |
| 5 | GPT-5.6 Sol | OpenAI | 83% |  | [MiMo-V2.6-Pro](https://handyai.news/modeldrop/mimo-v26-pro) |

19 more: https://handyai.news/api/benchmarks/osworld-verified

## MMLU-Pro

25 models from 9 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Claude Opus 4 | Anthropic | 87.3% |  | [GLM-4.5](https://handyai.news/modeldrop/glm-45) |
| 2 | Claude Sonnet 4.6 | Anthropic | 87% | max | [MAI-Thinking-1](https://handyai.news/modeldrop/mai-thinking-1) |
| 3 | Gemini 2.5 Pro | Google DeepMind | 86.2% |  | [GLM-4.5](https://handyai.news/modeldrop/glm-45) |
| 4 | o3 | OpenAI | 85.3% |  | [GLM-4.5](https://handyai.news/modeldrop/glm-45) |
| 5 | Gemma 4 | Google DeepMind | 85.2% |  | [Gemma 4](https://handyai.news/modeldrop/gemma-4) (own drop) |

20 more: https://handyai.news/api/benchmarks/mmlu-pro

## IFBench

24 models from 8 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Nemotron 3 Ultra | NVIDIA | 82.9% |  | [Nemotron 3.5 Super VL](https://handyai.news/modeldrop/nemotron-35-super-vl) |
| 2 | Qwen3.8-Max | Alibaba | 82.8% |  | [Qwen3.8-Max](https://handyai.news/modeldrop/qwen38-max) (own drop) |
| 3 | Inkling | Thinking Machines | 79.8% |  | [Inkling](https://handyai.news/modeldrop/inkling) (own drop) |
| 4 | Qwen3.5 397B | Alibaba | 78.2% |  | [Nemotron 3 Ultra](https://handyai.news/modeldrop/nemotron-3-ultra) |
| 5 | GLM-5.1 | Z.ai | 76.6% |  | [Nemotron 3 Ultra](https://handyai.news/modeldrop/nemotron-3-ultra) |

19 more: https://handyai.news/api/benchmarks/ifbench

## LMArena text Elo

23 models from 8 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Grok 4.1 | SpaceXAI | 1483 |  | [Grok 4.1](https://handyai.news/modeldrop/grok-41) (own drop) |
| 2 | Gemini 3 Flash | Google DeepMind | 1470 |  | [Gemini 3 Flash](https://handyai.news/modeldrop/gemini-3-flash) (own drop) |
| 3 | Gemini 2.5 Pro | Google DeepMind | 1452 |  | [Grok 4.1](https://handyai.news/modeldrop/grok-41) |
| 4 | GPT-5.1 | OpenAI | 1441 |  | [Grok 4.1](https://handyai.news/modeldrop/grok-41) |
| 5 | o3 | OpenAI | 1427 |  | [Claude Opus 4.1](https://handyai.news/modeldrop/claude-opus-41) |

18 more: https://handyai.news/api/benchmarks/lmarena-text-elo

## Terminal-Bench

Tasks in a real terminal (first edition). 21 models from 8 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Claude Sonnet 4.5 | Anthropic | 51% |  | [Kimi K2 Thinking](https://handyai.news/modeldrop/kimi-k2-thinking) |
| 2 | Kimi K2 Thinking | Moonshot AI | 47.1% | high | [Kimi K2 Thinking](https://handyai.news/modeldrop/kimi-k2-thinking) (own drop) |
| 3 | Claude Opus 4.1 | Anthropic | 46.5% |  | [Claude Sonnet 4.5](https://handyai.news/modeldrop/claude-sonnet-45) |
| 4 | MiniMax-M2 | MiniMax | 46.3% |  | [MiniMax-M2](https://handyai.news/modeldrop/minimax-m2) (own drop) |
| 5 | Kimi K2 0905 | Moonshot AI | 44.5% |  | [MiniMax-M2](https://handyai.news/modeldrop/minimax-m2) |

16 more: https://handyai.news/api/benchmarks/terminal-bench

## OSWorld 2

Using a computer: apps, files and the web. 16 models from 8 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Claude Sonnet 5.5 | Anthropic | 83.9% | max | [GPT-6.1 Terra](https://handyai.news/modeldrop/gpt-61-terra) |
| 2 | Claude Opus 5.5 | Anthropic | 81.8% |  | [Claude Sonnet 5.5](https://handyai.news/modeldrop/claude-sonnet-55) |
| 3 | Hy4 | Tencent | 78.9% |  | [Tencent Hunyuan Hy4 launch post](https://x.com/TencentHunyuan/status/2117870250838630539) (own drop) |
| 4 | GPT-6.1 Sol | OpenAI | 78% | max | [GPT-6.1 Terra](https://handyai.news/modeldrop/gpt-61-terra) |
| 5 | GPT-6.1 Terra | OpenAI | 74.2% | max | [GPT-6.1 Terra](https://handyai.news/modeldrop/gpt-61-terra) (own drop) |

11 more: https://handyai.news/api/benchmarks/osworld-2

## Harvey Legal Agent Benchmark

Legal work end to end. 21 models from 7 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Kimi K3.1 Pro | Moonshot AI | 23.7% | max | [Kimi K3.1 Pro](https://handyai.news/modeldrop/kimi-k31-pro) (own drop) |
| 2 | DeepSeek V4.1 Flash 1017 | DeepSeek | 21% | max | [Kimi K3.1 Pro](https://handyai.news/modeldrop/kimi-k31-pro) |
| 3 | Grok 4.7 | SpaceXAI | 19.6% | xhigh | [Grok 4.7](https://handyai.news/modeldrop/grok-47) (own drop) |
| 4 | Gemini 4 Argon | Google DeepMind | 19.6% | max | [Gemini 4 Argon](https://handyai.news/modeldrop/gemini-4-argon) (own drop) |
| 5 | Fable 5.1 | Anthropic | 19.1% | max | [Kimi K3.1 Pro](https://handyai.news/modeldrop/kimi-k31-pro) |

16 more: https://handyai.news/api/benchmarks/harvey-legal-agent-benchmark

## MCP Atlas

17 models from 6 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Muse Spark 1.2 | Meta | 90.3% | xhigh | [Muse Spark 1.2](https://handyai.news/modeldrop/muse-spark-12) (own drop) |
| 2 | Muse Spark Open | Meta | 89.2% | xhigh | [Muse Spark Open](https://handyai.news/modeldrop/muse-spark-open) (own drop) |
| 3 | Claude Opus 5 | Anthropic | 85.8% | xhigh | [Muse Spark 1.2](https://handyai.news/modeldrop/muse-spark-12) |
| 4 | Qwen3.8-2.4T-A95B | Alibaba | 84.5% | xhigh | [Muse Spark Open](https://handyai.news/modeldrop/muse-spark-open) |
| 5 | GLM-5.3 | Z.ai | 84.2% | max | [Muse Spark Open](https://handyai.news/modeldrop/muse-spark-open) |

12 more: https://handyai.news/api/benchmarks/mcp-atlas

## MMMLU

14 models from 6 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Claude Mythos Preview | Anthropic | 92.7% | max | [Claude Mythos Preview](https://handyai.news/modeldrop/claude-mythos-preview) (own drop) |
| 2 | Claude Opus 4.6 | Anthropic | 91.1% |  | [Claude Mythos Preview](https://handyai.news/modeldrop/claude-mythos-preview) |
| 3 | Claude Opus 4.1 | Anthropic | 89.5% |  | [Claude Opus 4.1](https://handyai.news/modeldrop/claude-opus-41) (own drop) |
| 4 | Claude Opus 4 | Anthropic | 88.8% |  | [Claude Opus 4.1](https://handyai.news/modeldrop/claude-opus-41) |
| 5 | o3 | OpenAI | 88.8% |  | [Claude Opus 4.1](https://handyai.news/modeldrop/claude-opus-41) |

9 more: https://handyai.news/api/benchmarks/mmmlu

## TAU-bench Retail

14 models from 6 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Claude Opus 4.1 | Anthropic | 82.4% |  | [Claude Opus 4.1](https://handyai.news/modeldrop/claude-opus-41) (own drop) |
| 2 | Claude Opus 4 | Anthropic | 81.4% |  | [Claude Opus 4.1](https://handyai.news/modeldrop/claude-opus-41) |
| 3 | Claude 3.7 Sonnet | Anthropic | 81.2% |  | [Claude Opus 4 & Sonnet 4](https://handyai.news/modeldrop/claude-4) |
| 4 | Claude Sonnet 4 | Anthropic | 80.5% |  | [Claude Opus 4.1](https://handyai.news/modeldrop/claude-opus-41) |
| 5 | GLM-4.5 | Z.ai | 79.7% |  | [GLM-4.5](https://handyai.news/modeldrop/glm-45) (own drop) |

9 more: https://handyai.news/api/benchmarks/tau-bench-retail

## Terminal-Bench Science

Scientific computing tasks in a terminal. 13 models from 6 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | GPT-6 Astra | OpenAI | 68.1% | max | [GPT-6.1 Terra](https://handyai.news/modeldrop/gpt-61-terra) |
| 2 | Claude Opus 5.5 | Anthropic | 63.3% | max | [GPT-6.1 Terra](https://handyai.news/modeldrop/gpt-61-terra) |
| 3 | GPT-6.1 Terra | OpenAI | 58.3% | max | [GPT-6.1 Terra](https://handyai.news/modeldrop/gpt-61-terra) (own drop) |
| 4 | GPT-6.1 Sol | OpenAI | 57% | max | [GPT-6.1 Terra](https://handyai.news/modeldrop/gpt-61-terra) |
| 5 | Fable 5.1 | Anthropic | 40% | max | [Kimi K3.1 Pro](https://handyai.news/modeldrop/kimi-k31-pro) |

8 more: https://handyai.news/api/benchmarks/terminal-bench-science

## Epoch Capabilities Index

93 models from 5 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | GPT-6 Astra | OpenAI | 169.1 |  | [Epoch Capabilities Index (2026-11-04)](https://epoch.ai/benchmarks/eci) |
| 2 | Claude Opus 5.5 | Anthropic | 168.4 |  | [Epoch Capabilities Index (2026-11-04)](https://epoch.ai/benchmarks/eci) |
| 3 | GPT-6.1 Terra | OpenAI | 168.1 |  | [Epoch Capabilities Index (2026-11-04)](https://epoch.ai/benchmarks/eci) |
| 4 | GPT-6.1 Sol | OpenAI | 167.8 |  | [Epoch Capabilities Index (2026-11-04)](https://epoch.ai/benchmarks/eci) |
| 5 | Claude Sonnet 5.5 | Anthropic | 166.7 |  | [Epoch Capabilities Index (2026-11-04)](https://epoch.ai/benchmarks/eci) |

88 more: https://handyai.news/api/benchmarks/epoch-capabilities-index

## AA-LCR

37 models from 5 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Kimi K3 | Moonshot AI | 88.7% | max | [Artificial Analysis (2026-11-04)](https://artificialanalysis.ai/evaluations) |
| 2 | MiMo-V2.6-Pro | Xiaomi | 86.3% |  | [Artificial Analysis (2026-11-04)](https://artificialanalysis.ai/evaluations) |
| 3 | Kimi K3.1 Pro | Moonshot AI | 86.3% |  | [Artificial Analysis (2026-11-04)](https://artificialanalysis.ai/evaluations) |
| 4 | Fable 5.1 | Anthropic | 85.3% | max | [Artificial Analysis (2026-11-04)](https://artificialanalysis.ai/evaluations) |
| 5 | Claude Opus 5.5 | Anthropic | 84.7% | max | [Artificial Analysis (2026-11-04)](https://artificialanalysis.ai/evaluations) |

32 more: https://handyai.news/api/benchmarks/aa-lcr

## ProgramBench

14 models from 5 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Claude Opus 5.5 | Anthropic | 65% | max | [Kimi K3.1 Pro](https://handyai.news/modeldrop/kimi-k31-pro) |
| 2 | Muse Wonder Full | Meta | 59% | max | [Muse Wonder](https://handyai.news/modeldrop/muse-wonder) |
| 3 | Fable 5.1 | Anthropic | 53.5% | max | [Kimi K3.1 Pro](https://handyai.news/modeldrop/kimi-k31-pro) |
| 4 | Muse Wonder | Meta | 50% | max | [Muse Wonder](https://handyai.news/modeldrop/muse-wonder) (own drop) |
| 5 | GPT-6 Astra | OpenAI | 50% | max | [Kimi K3.1 Pro](https://handyai.news/modeldrop/kimi-k31-pro) |

9 more: https://handyai.news/api/benchmarks/programbench

## CyberGym

13 models from 5 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | MAI-Cyber-1-Flash | Microsoft | 96% |  | [MAI-Cyber-1-Flash](https://handyai.news/modeldrop/mai-cyber-1-flash) (own drop) |
| 2 | GPT-5.5 Cyber | OpenAI | 85.6% |  | [MAI-Cyber-1-Flash](https://handyai.news/modeldrop/mai-cyber-1-flash) |
| 3 | GLM-5.3 | Z.ai | 84.5% |  | [MiniMax M3.1 Flash](https://handyai.news/modeldrop/minimax-m31-flash) |
| 4 | Mythos 5 | Anthropic | 83.8% |  | [GLM-5.3](https://handyai.news/modeldrop/glm-53) |
| 5 | GPT-5.6 Sol | OpenAI | 83.6% |  | [MiniMax M3.1 Flash](https://handyai.news/modeldrop/minimax-m31-flash) |

8 more: https://handyai.news/api/benchmarks/cybergym

## MATH-500

13 models from 5 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Nemotron Nano 2 | NVIDIA | 97.8% |  | [Nemotron Nano 2](https://handyai.news/modeldrop/nemotron-nano-2) (own drop) |
| 2 | DeepSeek R1 | DeepSeek | 97.3% |  | [Llama 3.1 Nemotron Ultra](https://handyai.news/modeldrop/llama-31-nemotron-ultra) |
| 3 | Llama 3.1 Nemotron Ultra | NVIDIA | 97% |  | [Llama 3.1 Nemotron Ultra](https://handyai.news/modeldrop/llama-31-nemotron-ultra) (own drop) |
| 4 | o1 | OpenAI | 96.4% |  | [DeepSeek R1](https://handyai.news/modeldrop/deepseek-r1) |
| 5 | Qwen3-8B | Alibaba | 96.3% |  | [Nemotron Nano 2](https://handyai.news/modeldrop/nemotron-nano-2) |

8 more: https://handyai.news/api/benchmarks/math-500

## SWE-bench Multilingual

13 models from 5 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | GPT-5.4 | OpenAI | 76.8% | high | [Composer 2](https://handyai.news/modeldrop/composer-2) |
| 2 | Claude Opus 4.6 | Anthropic | 75.8% | high | [Composer 2](https://handyai.news/modeldrop/composer-2) |
| 3 | Composer 2 | Cursor | 73.7% |  | [Composer 2](https://handyai.news/modeldrop/composer-2) (own drop) |
| 4 | Claude Sonnet 4.5 | Anthropic | 68% |  | [MiniMax-M2](https://handyai.news/modeldrop/minimax-m2) |
| 5 | MAI-Code-1-Flash | Microsoft | 65.5% |  | [MAI-Code-1-Flash](https://handyai.news/modeldrop/mai-code-1-flash) (own drop) |

8 more: https://handyai.news/api/benchmarks/swe-bench-multilingual

## Codeforces

12 models from 5 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | o3-pro | OpenAI | 2748 |  | [o3-pro](https://handyai.news/modeldrop/o3-pro) (own drop) |
| 2 | o4-mini | OpenAI | 2719 | high | [gpt-oss-120b & gpt-oss-20b](https://handyai.news/modeldrop/gpt-oss) |
| 3 | o3 | OpenAI | 2706 | high | [gpt-oss-120b & gpt-oss-20b](https://handyai.news/modeldrop/gpt-oss) |
| 4 | gpt-oss-120b | OpenAI | 2622 |  | [gpt-oss-120b & gpt-oss-20b](https://handyai.news/modeldrop/gpt-oss) (own drop) |
| 5 | gpt-oss-20b | OpenAI | 2516 |  | [gpt-oss-120b & gpt-oss-20b](https://handyai.news/modeldrop/gpt-oss) |

7 more: https://handyai.news/api/benchmarks/codeforces

## MMMU-Pro

16 models from 4 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Claude Opus 5.5 | Anthropic | 87.7% | max | [Kimi K3.1 Pro](https://handyai.news/modeldrop/kimi-k31-pro) |
| 2 | GPT-6 Astra | OpenAI | 86.9% | max | [Kimi K3.1 Pro](https://handyai.news/modeldrop/kimi-k31-pro) |
| 3 | Fable 5.1 | Anthropic | 86.3% | max | [Kimi K3.1 Pro](https://handyai.news/modeldrop/kimi-k31-pro) |
| 4 | Kimi K3.1 Pro | Moonshot AI | 84.1% | max | [Kimi K3.1 Pro](https://handyai.news/modeldrop/kimi-k31-pro) (own drop) |
| 5 | Gemini 3 Flash | Google DeepMind | 81.2% | xhigh | [Gemini 3 Flash](https://handyai.news/modeldrop/gemini-3-flash) (own drop) |

11 more: https://handyai.news/api/benchmarks/mmmu-pro

## NL2Repo

15 models from 4 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Claude Opus 5 | Anthropic | 75.3% |  | [MiniMax M3.1 Flash](https://handyai.news/modeldrop/minimax-m31-flash) |
| 2 | DeepSeek V4.1 Flash 1017 | DeepSeek | 68.2% |  | [MiniMax M3.1 Flash](https://handyai.news/modeldrop/minimax-m31-flash) |
| 3 | DeepSeek V4.1 Flash | DeepSeek | 65.4% |  | [MiniMax M3.1 Flash](https://handyai.news/modeldrop/minimax-m31-flash) |
| 4 | MiniMax M3.1 Flash | MiniMax | 61.2% |  | [MiniMax M3.1 Flash](https://handyai.news/modeldrop/minimax-m31-flash) (own drop) |
| 5 | Kimi K3 | Moonshot AI | 58% |  | [MiniMax M3.1 Flash](https://handyai.news/modeldrop/minimax-m31-flash) |

10 more: https://handyai.news/api/benchmarks/nl2repo

## CharXiv Reasoning

14 models from 4 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Muse Spark | Meta | 86.4% |  | [Muse Spark](https://handyai.news/modeldrop/muse-spark) (own drop) |
| 2 | GPT-5.6 Terra | OpenAI | 85.9% |  | [Gemini 3.7 Flash](https://handyai.news/modeldrop/gemini-37-flash) |
| 3 | Gemini 3.6 Flash | Google DeepMind | 85.2% | max | [Gemini 3.7 Flash](https://handyai.news/modeldrop/gemini-37-flash) |
| 4 | Gemini 3.7 Flash | Google DeepMind | 84.5% | high | [Gemini 3.7 Flash](https://handyai.news/modeldrop/gemini-37-flash) (own drop) |
| 5 | Qwen3.5-122B-A10B | Alibaba | 83.5% |  | [Nemotron 3.5 Super VL](https://handyai.news/modeldrop/nemotron-35-super-vl) |

9 more: https://handyai.news/api/benchmarks/charxiv-reasoning

## OSWorld 2 (offline subset)

Using a computer, on the 82 OSWorld 2.1 tasks that need no internet. 13 models from 4 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Claude Opus 5.5 | Anthropic | 87.2% | max | [Gemini 4 Carbon](https://handyai.news/modeldrop/gemini-4-carbon) |
| 2 | Claude Sonnet 5.5 | Anthropic | 83.9% |  | [Claude Haiku 5.5](https://handyai.news/modeldrop/claude-haiku-55) |
| 3 | Gemini 4 Carbon | Google DeepMind | 81.5% | max | [Gemini 4 Carbon](https://handyai.news/modeldrop/gemini-4-carbon) (own drop) |
| 4 | GPT-6 Astra | OpenAI | 80.9% | max | [Gemini 4 Carbon](https://handyai.news/modeldrop/gemini-4-carbon) |
| 5 | Fable 5.1 | Anthropic | 76.8% | max | [Gemini 4 Carbon](https://handyai.news/modeldrop/gemini-4-carbon) |

8 more: https://handyai.news/api/benchmarks/osworld-2-offline-subset

## AA-Omniscience hallucination rate

Lower is better. 12 models from 4 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Muse Wonder Full | Meta | 23% | max | [Muse Wonder](https://handyai.news/modeldrop/muse-wonder) |
| 2 | Muse Wonder | Meta | 26.5% | max | [Muse Wonder](https://handyai.news/modeldrop/muse-wonder) (own drop) |
| 3 | Grok 4.7 | SpaceXAI | 29% |  | [Grok 4.7](https://handyai.news/modeldrop/grok-47) (own drop) |
| 4 | Grok 4.6 | SpaceXAI | 34% |  | [Grok 4.7](https://handyai.news/modeldrop/grok-47) |
| 5 | GPT-6.1 Terra | OpenAI | 38% |  | [GPT-6.1 Terra](https://handyai.news/modeldrop/gpt-61-terra) (own drop) |

7 more: https://handyai.news/api/benchmarks/aa-omniscience-hallucination-rate

## AIME 2026

12 models from 4 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | GLM-5.2 | Z.ai | 99.2% |  | [Inkling](https://handyai.news/modeldrop/inkling) |
| 2 | Inkling | Thinking Machines | 97.1% |  | [Inkling](https://handyai.news/modeldrop/inkling) (own drop) |
| 3 | Kimi K2.6 | Moonshot AI | 96.4% |  | [Inkling](https://handyai.news/modeldrop/inkling) |
| 4 | Nanbeige4.3-33B | Nanbeige | 95.2% | max | [Nanbeige4.3-33B model card](https://huggingface.co/Nanbeige/Nanbeige4.3-33B) (own drop) |
| 5 | Muse Glimmer-30B | Meta | 94.7% | max | [Nanbeige4.3-33B model card](https://huggingface.co/Nanbeige/Nanbeige4.3-33B) |

7 more: https://handyai.news/api/benchmarks/aime-2026

## τ²-bench Retail

12 models from 4 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Claude Opus 4.1 | Anthropic | 86.8% |  | [Claude Sonnet 4.5](https://handyai.news/modeldrop/claude-sonnet-45) |
| 2 | Gemma 4 | Google DeepMind | 86.4% |  | [Gemma 4](https://handyai.news/modeldrop/gemma-4) (own drop) |
| 3 | Claude Sonnet 4.5 | Anthropic | 86.2% |  | [Claude Sonnet 4.5](https://handyai.news/modeldrop/claude-sonnet-45) (own drop) |
| 4 | Gemma 4 26B A4B | Google DeepMind | 85.5% |  | [Gemma 4](https://handyai.news/modeldrop/gemma-4) |
| 5 | Claude Sonnet 4 | Anthropic | 83.8% |  | [Claude Sonnet 4.5](https://handyai.news/modeldrop/claude-sonnet-45) |

7 more: https://handyai.news/api/benchmarks/bench-retail

## Agents' Last Exam

11 models from 4 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Kimi K3.1 Pro | Moonshot AI | 43.2% | max | [Kimi K3.1 Pro](https://handyai.news/modeldrop/kimi-k31-pro) (own drop) |
| 2 | Claude Opus 5.5 | Anthropic | 42.6% | max | [Kimi K3.1 Pro](https://handyai.news/modeldrop/kimi-k31-pro) |
| 3 | Hy4 | Tencent | 41% |  | [Tencent Hunyuan Hy4 launch post](https://x.com/TencentHunyuan/status/2117870250838630539) (own drop) |
| 4 | DeepSeek V4.1 Flash 1017 | DeepSeek | 35.4% |  | [DeepSeek V4.1 Flash 1017](https://handyai.news/modeldrop/deepseek-v41-flash-1017) (own drop) |
| 5 | GPT-6 Astra | OpenAI | 32.8% | max | [Kimi K3.1 Pro](https://handyai.news/modeldrop/kimi-k31-pro) |

6 more: https://handyai.news/api/benchmarks/agents-last-exam

## JobBench

11 models from 4 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Claude Opus 5 | Anthropic | 65.7% | max | [MiMo-V2.6-Pro](https://handyai.news/modeldrop/mimo-v26-pro) |
| 2 | Muse Spark 1.3 | Meta | 64.9% | max | [Muse Spark 1.3](https://handyai.news/modeldrop/muse-spark-13) (own drop) |
| 3 | MiMo-V2.6-Pro | Xiaomi | 62% |  | [MiMo-V2.6-Pro](https://handyai.news/modeldrop/mimo-v26-pro) (own drop) |
| 4 | Muse Spark Open | Meta | 61.8% | xhigh | [Muse Spark Open](https://handyai.news/modeldrop/muse-spark-open) (own drop) |
| 5 | GLM-5.3 | Z.ai | 61.4% | max | [Muse Spark Open](https://handyai.news/modeldrop/muse-spark-open) |

6 more: https://handyai.news/api/benchmarks/jobbench

## IFEval

9 models from 4 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | LFM3-1.7B | Liquid AI | 96.7% |  | [Liquid AI, "Introducing LFM3"](https://www.liquid.ai/blog/lfm3) (own drop) |
| 2 | Claude 3.7 Sonnet | Anthropic | 93.2% |  | [Claude 3.7 Sonnet](https://handyai.news/modeldrop/claude-37-sonnet) (own drop) |
| 3 | Nemotron Nano 2 | NVIDIA | 90.3% |  | [Nemotron Nano 2](https://handyai.news/modeldrop/nemotron-nano-2) (own drop) |
| 4 | Claude 3.5 Sonnet | Anthropic | 90.2% |  | [Claude 3.7 Sonnet](https://handyai.news/modeldrop/claude-37-sonnet) |
| 5 | Qwen3-8B | Alibaba | 89.4% |  | [Nemotron Nano 2](https://handyai.news/modeldrop/nemotron-nano-2) |

4 more: https://handyai.news/api/benchmarks/ifeval

## MMLU

9 models from 4 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | o3 | OpenAI | 93.4% |  | [gpt-oss-120b & gpt-oss-20b](https://handyai.news/modeldrop/gpt-oss) |
| 2 | o4-mini | OpenAI | 93% |  | [gpt-oss-120b & gpt-oss-20b](https://handyai.news/modeldrop/gpt-oss) |
| 3 | o1 | OpenAI | 91.8% |  | [o1](https://handyai.news/modeldrop/o1) (own drop) |
| 4 | o1-preview | OpenAI | 90.8% |  | [o1](https://handyai.news/modeldrop/o1) |
| 5 | gpt-oss-120b | OpenAI | 90% |  | [gpt-oss-120b & gpt-oss-20b](https://handyai.news/modeldrop/gpt-oss) (own drop) |

4 more: https://handyai.news/api/benchmarks/mmlu

## Terminal-Bench 4.0 (standard harness)

Terminal-Bench run by a third party on one shared harness. 29 models from 3 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Kimi K3.1 Pro | Moonshot AI | 67.2% |  | [Artificial Analysis (2026-11-04)](https://artificialanalysis.ai/evaluations) |
| 2 | Gemini 4 Carbon | Google DeepMind | 65.2% | high | [Artificial Analysis (2026-11-04)](https://artificialanalysis.ai/evaluations) |
| 3 | Claude Sonnet 5.5 | Anthropic | 63.6% | max | [Artificial Analysis (2026-11-04)](https://artificialanalysis.ai/evaluations) |
| 4 | Claude Opus 5.5 | Anthropic | 59.6% | max | [Artificial Analysis (2026-11-04)](https://artificialanalysis.ai/evaluations) |
| 5 | GPT-6 Astra | OpenAI | 59.1% | max | [Artificial Analysis (2026-11-04)](https://artificialanalysis.ai/evaluations) |

24 more: https://handyai.news/api/benchmarks/terminal-bench-4-0-standard-harness

## GDP.pdf

Work products built from PDFs (Surge AI). 13 models from 3 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Kimi K3.1 Pro | Moonshot AI | 34.6% | max | [Kimi K3.1 Pro](https://handyai.news/modeldrop/kimi-k31-pro) (own drop) |
| 2 | GPT-6 Astra | OpenAI | 34.2% | max | [Kimi K3.1 Pro](https://handyai.news/modeldrop/kimi-k31-pro) |
| 3 | Gemini 3.7 Flash | Google DeepMind | 34% | high | [Gemini 3.7 Flash](https://handyai.news/modeldrop/gemini-37-flash) (own drop) |
| 4 | GPT-6.1 Sol | OpenAI | 32% | high | [GPT-6.1 Sol](https://handyai.news/modeldrop/gpt-61-sol) (own drop) |
| 5 | Claude Opus 5.5 | Anthropic | 30.6% | max | [Kimi K3.1 Pro](https://handyai.news/modeldrop/kimi-k31-pro) |

8 more: https://handyai.news/api/benchmarks/gdp-pdf

## GPQA

13 models from 3 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | GPT-5.2 | OpenAI | 92.4% |  | [Qwen3.5](https://handyai.news/modeldrop/qwen35) |
| 2 | Gemini 3 Pro | Google DeepMind | 91.9% |  | [Qwen3.5](https://handyai.news/modeldrop/qwen35) |
| 3 | Qwen3.5 | Alibaba | 88.4% |  | [Qwen3.5](https://handyai.news/modeldrop/qwen35) (own drop) |
| 4 | Grok 4 | SpaceXAI | 87.7% |  | [GLM-4.5](https://handyai.news/modeldrop/glm-45) |
| 5 | Kimi K2.5 | Moonshot AI | 87.6% |  | [Qwen3.5](https://handyai.news/modeldrop/qwen35) |

8 more: https://handyai.news/api/benchmarks/gpqa

## ExploitBench

11 models from 3 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | GPT-6 Astra | OpenAI | 100% | max | [Kimi K3.1 Pro](https://handyai.news/modeldrop/kimi-k31-pro) |
| 2 | Claude Opus 5.5 | Anthropic | 91% | max | [Kimi K3.1 Pro](https://handyai.news/modeldrop/kimi-k31-pro) |
| 3 | Fable 5.1 | Anthropic | 83% | max | [Kimi K3.1 Pro](https://handyai.news/modeldrop/kimi-k31-pro) |
| 4 | Mythos 5 | Anthropic | 78% |  | [GLM-5.3](https://handyai.news/modeldrop/glm-53) |
| 5 | GPT-5.6 Sol | OpenAI | 76.5% |  | [GLM-5.3](https://handyai.news/modeldrop/glm-53) |

6 more: https://handyai.news/api/benchmarks/exploitbench

## Terminal-Bench 3.0

11 models from 3 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Claude Opus 5 | Anthropic | 43.5% |  | [Grok 4.6](https://handyai.news/modeldrop/grok-46) |
| 2 | GPT-5.6 Sol | OpenAI | 34.6% | max | [GLM-5.3](https://handyai.news/modeldrop/glm-53) |
| 3 | Fable 5 | Anthropic | 33.7% |  | [GLM-5.3](https://handyai.news/modeldrop/glm-53) |
| 4 | GLM-5.3 | Z.ai | 28.3% |  | [GLM-5.3](https://handyai.news/modeldrop/glm-53) (own drop) |
| 5 | Grok 4.6 | SpaceXAI | 26% | high | [Grok 4.6](https://handyai.news/modeldrop/grok-46) (own drop) |

6 more: https://handyai.news/api/benchmarks/terminal-bench-3-0

## ZeroBench (with tools)

11 models from 3 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Kimi K3.1 Pro | Moonshot AI | 63% | max | [Kimi K3.1 Pro](https://handyai.news/modeldrop/kimi-k31-pro) (own drop) |
| 2 | Claude Opus 5.5 | Anthropic | 61% | max | [Kimi K3.1 Pro](https://handyai.news/modeldrop/kimi-k31-pro) |
| 3 | Fable 5.1 | Anthropic | 59% | max | [Kimi K3.1 Pro](https://handyai.news/modeldrop/kimi-k31-pro) |
| 4 | GPT-6 Astra | OpenAI | 58% | max | [Kimi K3.1 Pro](https://handyai.news/modeldrop/kimi-k31-pro) |
| 5 | GPT-5.6 Sol | OpenAI | 53% |  | [DeepSeek V4.1 Flash 1017](https://handyai.news/modeldrop/deepseek-v41-flash-1017) |

6 more: https://handyai.news/api/benchmarks/zerobench-with-tools

## BabyVision (with tools)

10 models from 3 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Claude Opus 5.5 | Anthropic | 97.7% | max | [Kimi K3.1 Pro](https://handyai.news/modeldrop/kimi-k31-pro) |
| 2 | GPT-6 Astra | OpenAI | 96.6% | max | [Kimi K3.1 Pro](https://handyai.news/modeldrop/kimi-k31-pro) |
| 3 | Kimi K3.1 Pro | Moonshot AI | 95.4% | max | [Kimi K3.1 Pro](https://handyai.news/modeldrop/kimi-k31-pro) (own drop) |
| 4 | Fable 5.1 | Anthropic | 95.1% | max | [Kimi K3.1 Pro](https://handyai.news/modeldrop/kimi-k31-pro) |
| 5 | Claude Opus 5 | Anthropic | 94.1% |  | [DeepSeek V4.1 Flash 1017](https://handyai.news/modeldrop/deepseek-v41-flash-1017) |

5 more: https://handyai.news/api/benchmarks/babyvision-with-tools

## FrontierCode v1.1

Hard, long coding tasks. 9 models from 3 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Claude Opus 5.5 | Anthropic | 54.4% |  | [Claude Sonnet 5.5](https://handyai.news/modeldrop/claude-sonnet-55) |
| 2 | GPT-6 Astra | OpenAI | 53.3% |  | [Claude Opus 5.5](https://handyai.news/modeldrop/claude-opus-55) |
| 3 | Claude Sonnet 5.5 | Anthropic | 52.1% | xhigh | [Claude Haiku 5.5](https://handyai.news/modeldrop/claude-haiku-55) |
| 4 | Fable 5.1 | Anthropic | 50.3% |  | [Claude Opus 5.5](https://handyai.news/modeldrop/claude-opus-55) |
| 5 | GPT-6 Sol | OpenAI | 49.3% |  | [Claude Sonnet 5.5](https://handyai.news/modeldrop/claude-sonnet-55) |

4 more: https://handyai.news/api/benchmarks/frontiercode-v1-1

## SciCode

6 models from 3 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Ling-3.1-flash-1019 | Ant Group | 57% |  | [Ling-3.1-flash-1019 model card](https://huggingface.co/inclusionAI/Ling-3.1-flash-1019) (own drop) |
| 2 | Nanbeige4.3-33B | Nanbeige | 52.6% | max | [Nanbeige4.3-33B model card](https://huggingface.co/Nanbeige/Nanbeige4.3-33B) (own drop) |
| 3 | Qwen3.8 | Alibaba | 46.7% | max | [Nanbeige4.3-33B model card](https://huggingface.co/Nanbeige/Nanbeige4.3-33B) |
| 4 | Gemma 4 | Google DeepMind | 45.5% | max | [Nanbeige4.3-33B model card](https://huggingface.co/Nanbeige/Nanbeige4.3-33B) |
| 5 | Muse Glimmer-30B | Meta | 44.9% | max | [Nanbeige4.3-33B model card](https://huggingface.co/Nanbeige/Nanbeige4.3-33B) |

1 more: https://handyai.news/api/benchmarks/scicode

## Tau3-Bench

6 models from 3 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | LFM3-1.7B | Liquid AI | 68% |  | [Liquid AI, "Introducing LFM3"](https://www.liquid.ai/blog/lfm3) (own drop) |
| 2 | Clover-35B-A3B-Preview | DiffuSpace | 65.3% |  | [DiffuSpace, "Clover-35B-A3B-Preview"](https://diffuspace.ai/blog/clover-35b/) (own drop) |
| 3 | Nanbeige4.3-33B | Nanbeige | 48.8% | max | [Nanbeige4.3-33B model card](https://huggingface.co/Nanbeige/Nanbeige4.3-33B) (own drop) |
| 4 | Qwen3.8 | Alibaba | 48.5% | max | [Nanbeige4.3-33B model card](https://huggingface.co/Nanbeige/Nanbeige4.3-33B) |
| 5 | Muse Glimmer-30B | Meta | 33.7% | max | [Nanbeige4.3-33B model card](https://huggingface.co/Nanbeige/Nanbeige4.3-33B) |

1 more: https://handyai.news/api/benchmarks/tau3-bench

## GDPval

Work products judged against industry experts (win or tie rate). 4 models from 3 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | GPT-5.4 | OpenAI | 83% |  | [GPT-5.4](https://handyai.news/modeldrop/gpt-54) (own drop) |
| 2 | GPT-5.2 | OpenAI | 70.9% | xhigh | [GPT-5.4](https://handyai.news/modeldrop/gpt-54) |
| 3 | GPT-5.3-Codex | OpenAI | 70.9% |  | [GPT-5.4](https://handyai.news/modeldrop/gpt-54) |
| 4 | GPT-5 | OpenAI | 38.8% | xhigh | [GPT-5.2](https://handyai.news/modeldrop/gpt-52) |

## AA-Briefcase v1.1

29 models from 2 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Claude Sonnet 5.5 | Anthropic | 1813 Elo | max | [Artificial Analysis (2026-11-04)](https://artificialanalysis.ai/evaluations) |
| 2 | Claude Opus 5.5 | Anthropic | 1800 Elo | max | [Artificial Analysis (2026-11-04)](https://artificialanalysis.ai/evaluations) |
| 3 | Muse Wonder Full | Meta | 1729 Elo | max | [Artificial Analysis (2026-11-04)](https://artificialanalysis.ai/evaluations) |
| 4 | Kimi K3.1 Pro | Moonshot AI | 1675 Elo |  | [Artificial Analysis (2026-11-04)](https://artificialanalysis.ai/evaluations) |
| 5 | Fable 5.1 | Anthropic | 1669 Elo | max | [Artificial Analysis (2026-11-04)](https://artificialanalysis.ai/evaluations) |

24 more: https://handyai.news/api/benchmarks/aa-briefcase-v1-1

## AutomationBench-AA

Business automations, run by Artificial Analysis. 27 models from 2 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Gemini 4 Carbon | Google DeepMind | 77.8% | high | [Artificial Analysis (2026-11-04)](https://artificialanalysis.ai/evaluations) |
| 2 | Gemini 4 Argon | Google DeepMind | 77.5% | high | [Artificial Analysis (2026-11-04)](https://artificialanalysis.ai/evaluations) |
| 3 | Kimi K3.1 Pro | Moonshot AI | 73.3% |  | [Artificial Analysis (2026-11-04)](https://artificialanalysis.ai/evaluations) |
| 4 | MiniMax M3.1 Flash | MiniMax | 72.5% |  | [Artificial Analysis (2026-11-04)](https://artificialanalysis.ai/evaluations) |
| 5 | Claude Sonnet 5.5 | Anthropic | 71.8% | max | [Artificial Analysis (2026-11-04)](https://artificialanalysis.ai/evaluations) |

22 more: https://handyai.news/api/benchmarks/automationbench-aa

## APEX-Agents

11 models from 2 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Claude Opus 5.5 | Anthropic | 73.5% | max | [Kimi K3.1 Pro](https://handyai.news/modeldrop/kimi-k31-pro) |
| 2 | Fable 5.1 | Anthropic | 68.6% | max | [Kimi K3.1 Pro](https://handyai.news/modeldrop/kimi-k31-pro) |
| 3 | Kimi K3.1 Pro | Moonshot AI | 67.2% | max | [Kimi K3.1 Pro](https://handyai.news/modeldrop/kimi-k31-pro) (own drop) |
| 4 | GPT-6 Astra | OpenAI | 64.7% | max | [Kimi K3.1 Pro](https://handyai.news/modeldrop/kimi-k31-pro) |
| 5 | DeepSeek V4.1 Flash 1017 | DeepSeek | 63.4% | max | [Kimi K3.1 Pro](https://handyai.news/modeldrop/kimi-k31-pro) |

6 more: https://handyai.news/api/benchmarks/apex-agents

## Agents' Last Exam v1

9 models from 2 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Claude Opus 5.5 | Anthropic | 63.2% | max | [GPT-6.1 Terra](https://handyai.news/modeldrop/gpt-61-terra) |
| 2 | GPT-6.1 Terra | OpenAI | 60% | max | [GPT-6.1 Terra](https://handyai.news/modeldrop/gpt-61-terra) (own drop) |
| 3 | GPT-6 Astra | OpenAI | 59.3% | max | [GPT-6.1 Terra](https://handyai.news/modeldrop/gpt-61-terra) |
| 4 | GPT-6.1 Sol | OpenAI | 58.6% | max | [GPT-6.1 Terra](https://handyai.news/modeldrop/gpt-61-terra) |
| 5 | GPT-6 Sol | OpenAI | 56.4% | max | [GPT-6 Sol & GPT-6 Luna](https://handyai.news/modeldrop/gpt-6-sol-and-gpt-6-luna) (own drop) |

4 more: https://handyai.news/api/benchmarks/agents-last-exam-v1

## Chartography (with tools)

9 models from 2 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | GPT-6 Astra | OpenAI | 91% | max | [Kimi K3.1 Pro](https://handyai.news/modeldrop/kimi-k31-pro) |
| 2 | Kimi K3.1 Pro | Moonshot AI | 90.2% | max | [Kimi K3.1 Pro](https://handyai.news/modeldrop/kimi-k31-pro) (own drop) |
| 3 | Claude Opus 5.5 | Anthropic | 89% | max | [Kimi K3.1 Pro](https://handyai.news/modeldrop/kimi-k31-pro) |
| 4 | Fable 5.1 | Anthropic | 88.4% | max | [Kimi K3.1 Pro](https://handyai.news/modeldrop/kimi-k31-pro) |
| 5 | Claude Opus 5 | Anthropic | 84% |  | [DeepSeek V4.1 Flash 1017](https://handyai.news/modeldrop/deepseek-v41-flash-1017) |

4 more: https://handyai.news/api/benchmarks/chartography-with-tools

## PinchBench

9 models from 2 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Kimi K2.6 | Moonshot AI | 90.2% |  | [Nemotron 3 Ultra](https://handyai.news/modeldrop/nemotron-3-ultra) |
| 2 | Nemotron 3 Ultra | NVIDIA | 90% |  | [Nemotron 3 Ultra](https://handyai.news/modeldrop/nemotron-3-ultra) (own drop) |
| 3 | Qwen3.6 35B A3B | Alibaba | 88.1% |  | [Nemotron 3.5 Lightning](https://handyai.news/modeldrop/nemotron-35-lightning) |
| 4 | Qwen3.5 397B | Alibaba | 86.6% |  | [Nemotron 3 Ultra](https://handyai.news/modeldrop/nemotron-3-ultra) |
| 5 | Nemotron 3.5 Lightning | NVIDIA | 85.4% |  | [Nemotron 3.5 Lightning](https://handyai.news/modeldrop/nemotron-35-lightning) (own drop) |

4 more: https://handyai.news/api/benchmarks/pinchbench

## Agents' Last Exam (CLI)

8 models from 2 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Kimi K3.1 Pro | Moonshot AI | 44.2% | max | [Kimi K3.1 Pro](https://handyai.news/modeldrop/kimi-k31-pro) (own drop) |
| 2 | Claude Opus 5.5 | Anthropic | 40.7% | max | [Kimi K3.1 Pro](https://handyai.news/modeldrop/kimi-k31-pro) |
| 3 | GPT-6 Astra | OpenAI | 38.8% | max | [Kimi K3.1 Pro](https://handyai.news/modeldrop/kimi-k31-pro) |
| 4 | DeepSeek V4.1 Flash 1017 | DeepSeek | 36% | max | [Kimi K3.1 Pro](https://handyai.news/modeldrop/kimi-k31-pro) |
| 5 | GPT-5.6 Sol | OpenAI | 28.6% |  | [GLM-5.3](https://handyai.news/modeldrop/glm-53) |

3 more: https://handyai.news/api/benchmarks/agents-last-exam-cli

## HealthBench Professional

8 models from 2 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Mythos 5 | Anthropic | 66% |  | [Fable 5 / Mythos 5](https://handyai.news/modeldrop/fable-5-mythos-5) (own drop) |
| 2 | Claude Mythos Preview | Anthropic | 64.7% |  | [Fable 5 / Mythos 5](https://handyai.news/modeldrop/fable-5-mythos-5) |
| 3 | Fable 5.1 | Anthropic | 62.1% | max | [Grok 4.7](https://handyai.news/modeldrop/grok-47) |
| 4 | GPT-5.6 Sol | OpenAI | 60.5% | max | [Grok 4.7](https://handyai.news/modeldrop/grok-47) |
| 5 | Claude Opus 4.8 | Anthropic | 56.9% |  | [Fable 5 / Mythos 5](https://handyai.news/modeldrop/fable-5-mythos-5) |

3 more: https://handyai.news/api/benchmarks/healthbench-professional

## OSWorld

8 models from 2 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Claude Opus 5.5 | Anthropic | 81.8% | max | [Kimi K3.1 Pro](https://handyai.news/modeldrop/kimi-k31-pro) |
| 2 | Fable 5.1 | Anthropic | 80.7% | max | [Kimi K3.1 Pro](https://handyai.news/modeldrop/kimi-k31-pro) |
| 3 | Kimi K3.1 Pro | Moonshot AI | 79.6% | max | [Kimi K3.1 Pro](https://handyai.news/modeldrop/kimi-k31-pro) (own drop) |
| 4 | GPT-6 Astra | OpenAI | 75.8% | max | [Kimi K3.1 Pro](https://handyai.news/modeldrop/kimi-k31-pro) |
| 5 | DeepSeek V4.1 Flash 1017 | DeepSeek | 69.3% | max | [Kimi K3.1 Pro](https://handyai.news/modeldrop/kimi-k31-pro) |

3 more: https://handyai.news/api/benchmarks/osworld

## SWE-Marathon

8 models from 2 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Muse Wonder Full | Meta | 57.5% | xhigh | [Muse Wonder](https://handyai.news/modeldrop/muse-wonder) |
| 2 | Muse Wonder | Meta | 53.1% | max | [Muse Wonder](https://handyai.news/modeldrop/muse-wonder) (own drop) |
| 3 | Kimi K3.1 Pro | Moonshot AI | 51.7% | max | [Kimi K3.1 Pro](https://handyai.news/modeldrop/kimi-k31-pro) (own drop) |
| 4 | Claude Opus 5.5 | Anthropic | 46.3% | max | [Kimi K3.1 Pro](https://handyai.news/modeldrop/kimi-k31-pro) |
| 5 | Fable 5.1 | Anthropic | 45.6% | max | [Kimi K3.1 Pro](https://handyai.news/modeldrop/kimi-k31-pro) |

3 more: https://handyai.news/api/benchmarks/swe-marathon

## Toolathlon-Verified

8 models from 2 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Claude Opus 5 | Anthropic | 80.6% |  | [MiMo-V2.6-Pro](https://handyai.news/modeldrop/mimo-v26-pro) |
| 2 | MiMo-V2.6-Pro | Xiaomi | 76.9% |  | [MiMo-V2.6-Pro](https://handyai.news/modeldrop/mimo-v26-pro) (own drop) |
| 3 | Kimi K3 | Moonshot AI | 76.5% |  | [MiMo-V2.6-Pro](https://handyai.news/modeldrop/mimo-v26-pro) |
| 4 | GPT-5.6 Sol | OpenAI | 74.9% |  | [MiMo-V2.6-Pro](https://handyai.news/modeldrop/mimo-v26-pro) |
| 5 | Nanbeige4.3-33B | Nanbeige | 70.4% | max | [Nanbeige4.3-33B model card](https://huggingface.co/Nanbeige/Nanbeige4.3-33B) (own drop) |

3 more: https://handyai.news/api/benchmarks/toolathlon-verified

## Vals Index

8 models from 2 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Gemini 4 Argon | Google DeepMind | 68.9% | max | [Gemini 4 Argon](https://handyai.news/modeldrop/gemini-4-argon) (own drop) |
| 2 | Claude Opus 5.5 | Anthropic | 67% |  | [Gemini 4 Argon](https://handyai.news/modeldrop/gemini-4-argon) |
| 3 | Fable 5.1 | Anthropic | 65.8% |  | [Gemini 4 Argon](https://handyai.news/modeldrop/gemini-4-argon) |
| 4 | GPT-6 Astra | OpenAI | 63.1% |  | [Gemini 4 Argon](https://handyai.news/modeldrop/gemini-4-argon) |
| 5 | GLM-5.3 | Z.ai | 53.5% |  | [Mistral Large 4](https://handyai.news/modeldrop/mistral-large-4) |

3 more: https://handyai.news/api/benchmarks/vals-index

## Video-MMMU

8 models from 2 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Gemini 3 Pro | Google DeepMind | 87.6% |  | [Kimi K2.5](https://handyai.news/modeldrop/kimi-k25) |
| 2 | Kimi K2.5 | Moonshot AI | 86.6% |  | [Kimi K2.5](https://handyai.news/modeldrop/kimi-k25) (own drop) |
| 3 | GPT-5.2 | OpenAI | 85.9% | xhigh | [Kimi K2.5](https://handyai.news/modeldrop/kimi-k25) |
| 4 | Claude Opus 4.5 | Anthropic | 84.4% |  | [Kimi K2.5](https://handyai.news/modeldrop/kimi-k25) |
| 5 | Gemini 2.5 Pro | Google DeepMind | 83.6% |  | [Gemini 3 Pro](https://handyai.news/modeldrop/gemini-3-pro) |

3 more: https://handyai.news/api/benchmarks/video-mmmu

## BFCL v3

7 models from 2 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Qwen3 | Alibaba | 70.8% |  | [Qwen3](https://handyai.news/modeldrop/qwen3) (own drop) |
| 2 | o1 | OpenAI | 67.8% |  | [Qwen3](https://handyai.news/modeldrop/qwen3) |
| 3 | Nemotron Nano 2 | NVIDIA | 66.9% |  | [Nemotron Nano 2](https://handyai.news/modeldrop/nemotron-nano-2) (own drop) |
| 4 | Qwen3-8B | Alibaba | 66.3% |  | [Nemotron Nano 2](https://handyai.news/modeldrop/nemotron-nano-2) |
| 5 | o3-mini | OpenAI | 64.6% | medium | [Qwen3](https://handyai.news/modeldrop/qwen3) |

2 more: https://handyai.news/api/benchmarks/bfcl-v3

## DeepSearchQA

7 models from 2 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | GPT-5.6 Sol | OpenAI | 93.1% | max | [Muse Spark 1.3](https://handyai.news/modeldrop/muse-spark-13) |
| 2 | Claude Opus 5 | Anthropic | 90.4% | max | [Muse Spark 1.3](https://handyai.news/modeldrop/muse-spark-13) |
| 3 | Muse Spark 1.3 | Meta | 90.3% | max | [Muse Spark 1.3](https://handyai.news/modeldrop/muse-spark-13) (own drop) |
| 4 | Muse Spark | Meta | 74.8% |  | [Muse Spark](https://handyai.news/modeldrop/muse-spark) (own drop) |
| 5 | Claude Opus 4.6 | Anthropic | 73.7% | max | [Muse Spark](https://handyai.news/modeldrop/muse-spark) |

2 more: https://handyai.news/api/benchmarks/deepsearchqa

## HLE-Verified

7 models from 2 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Gemini 3.8 Flash | Google DeepMind | 54.9% | max | [Gemini 3.8 Flash](https://handyai.news/modeldrop/gemini-38-flash) (own drop) |
| 2 | GPT-5.6 Sol | OpenAI | 54.5% | max | [Gemini 3.8 Flash](https://handyai.news/modeldrop/gemini-38-flash) |
| 3 | Claude Opus 5 | Anthropic | 54.4% | max | [Gemini 3.8 Flash](https://handyai.news/modeldrop/gemini-38-flash) |
| 4 | Gemini 3.7 Flash | Google DeepMind | 53.6% | high | [Gemini 3.7 Flash](https://handyai.news/modeldrop/gemini-37-flash) (own drop) |
| 5 | Gemini 3.6 Flash | Google DeepMind | 51.2% |  | [Gemini 3.7 Flash](https://handyai.news/modeldrop/gemini-37-flash) |

2 more: https://handyai.news/api/benchmarks/hle-verified

## HMMT 2025

7 models from 2 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Grok 4 Fast | SpaceXAI | 93.3% |  | [Grok 4 Fast](https://handyai.news/modeldrop/grok-4-fast) (own drop) |
| 2 | GPT-5 | OpenAI | 93.3% | high | [Grok 4 Fast](https://handyai.news/modeldrop/grok-4-fast) |
| 3 | Grok 4 | SpaceXAI | 90% |  | [Grok 4 Fast](https://handyai.news/modeldrop/grok-4-fast) |
| 4 | GPT-5 mini | OpenAI | 87.8% | high | [Grok 4 Fast](https://handyai.news/modeldrop/grok-4-fast) |
| 5 | Gemini 2.5 Pro | Google DeepMind | 82.5% |  | [Grok 4](https://handyai.news/modeldrop/grok-4) |

2 more: https://handyai.news/api/benchmarks/hmmt-2025

## SimpleQA Verified

7 models from 2 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Gemini 3 Pro | Google DeepMind | 72.1% |  | [Gemini 3 Pro](https://handyai.news/modeldrop/gemini-3-pro) (own drop) |
| 2 | Gemini 3 Flash | Google DeepMind | 68.7% | xhigh | [Gemini 3 Flash](https://handyai.news/modeldrop/gemini-3-flash) (own drop) |
| 3 | Gemini 2.5 Pro | Google DeepMind | 54.5% |  | [Gemini 3 Pro](https://handyai.news/modeldrop/gemini-3-pro) |
| 4 | GPT-5.2 | OpenAI | 38% | xhigh | [Gemini 3 Flash](https://handyai.news/modeldrop/gemini-3-flash) |
| 5 | GPT-5.1 | OpenAI | 34.9% |  | [Gemini 3 Pro](https://handyai.news/modeldrop/gemini-3-pro) |

2 more: https://handyai.news/api/benchmarks/simpleqa-verified

## Terminal-Bench 2.1 (standard harness)

7 models from 2 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Claude Opus 5 | Anthropic | 89.1% |  | [Muse Spark 1.3](https://handyai.news/modeldrop/muse-spark-13) |
| 2 | GPT-5.6 Terra | OpenAI | 88% |  | [Muse Spark 1.2](https://handyai.news/modeldrop/muse-spark-12) |
| 3 | GPT-5.6 Sol | OpenAI | 88% |  | [Muse Spark 1.3](https://handyai.news/modeldrop/muse-spark-13) |
| 4 | Gemini 3.8 Flash | Google DeepMind | 87.6% | high | [Muse Spark 1.3](https://handyai.news/modeldrop/muse-spark-13) |
| 5 | Muse Spark 1.3 | Meta | 85.4% | xhigh | [Muse Spark 1.3](https://handyai.news/modeldrop/muse-spark-13) (own drop) |

2 more: https://handyai.news/api/benchmarks/terminal-bench-2-1-standard-harness

## AA-Briefcase

6 models from 2 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Claude Opus 5 | Anthropic | 1715 Elo |  | [Grok 4.6](https://handyai.news/modeldrop/grok-46) |
| 2 | Grok 4.6 | SpaceXAI | 1577 Elo | high | [Grok 4.6](https://handyai.news/modeldrop/grok-46) (own drop) |
| 3 | Fable 5 | Anthropic | 1574 Elo | max | [Grok 4.6](https://handyai.news/modeldrop/grok-46) |
| 4 | Ling-3.1-flash-1019 | Ant Group | 1524 Elo |  | [Ling-3.1-flash-1019 model card](https://huggingface.co/inclusionAI/Ling-3.1-flash-1019) (own drop) |
| 5 | GPT-5.6 Sol | OpenAI | 1502 Elo | max | [Grok 4.6](https://handyai.news/modeldrop/grok-46) |

1 more: https://handyai.news/api/benchmarks/aa-briefcase

## CursorBench

6 models from 2 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | GPT-5.4 | OpenAI | 63.9% | high | [Composer 2](https://handyai.news/modeldrop/composer-2) |
| 2 | Composer 2 | Cursor | 61.3% |  | [Composer 2](https://handyai.news/modeldrop/composer-2) (own drop) |
| 3 | GPT-5.3-Codex | OpenAI | 59.1% |  | [Composer 2](https://handyai.news/modeldrop/composer-2) |
| 4 | Claude Opus 4.6 | Anthropic | 58.2% | high | [Composer 2](https://handyai.news/modeldrop/composer-2) |
| 5 | Composer | Cursor | 38% |  | [Composer](https://handyai.news/modeldrop/composer) (own drop) |

1 more: https://handyai.news/api/benchmarks/cursorbench

## CursorBench 4.0

Long coding tasks inside Cursor. 6 models from 2 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Claude Opus 5.5 | Anthropic | 57.8% |  | [Claude Opus 5.5](https://handyai.news/modeldrop/claude-opus-55) (own drop) |
| 2 | Fable 5.1 | Anthropic | 51.8% | max | [Claude Opus 5.5](https://handyai.news/modeldrop/claude-opus-55) |
| 3 | Claude Opus 5 | Anthropic | 46.6% |  | [Claude Opus 5.5](https://handyai.news/modeldrop/claude-opus-55) |
| 4 | Grok 4.7 | SpaceXAI | 46.3% | xhigh | [Grok 4.7](https://handyai.news/modeldrop/grok-47) (own drop) |
| 5 | GPT-5.6 Sol | OpenAI | 41.7% | max | [Grok 4.7](https://handyai.news/modeldrop/grok-47) |

1 more: https://handyai.news/api/benchmarks/cursorbench-4-0

## Multi-IF

6 models from 2 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | LFM3-1.7B | Liquid AI | 89.1% |  | [Liquid AI, "Introducing LFM3"](https://www.liquid.ai/blog/lfm3) (own drop) |
| 2 | Gemini 2.5 Pro | Google DeepMind | 77.8% |  | [Qwen3](https://handyai.news/modeldrop/qwen3) |
| 3 | Qwen3 | Alibaba | 71.9% |  | [Qwen3](https://handyai.news/modeldrop/qwen3) (own drop) |
| 4 | DeepSeek R1 | DeepSeek | 67.7% |  | [Qwen3](https://handyai.news/modeldrop/qwen3) |
| 5 | o1 | OpenAI | 48.8% |  | [Qwen3](https://handyai.news/modeldrop/qwen3) |

1 more: https://handyai.news/api/benchmarks/multi-if

## SWE-Atlas-QnA

6 models from 2 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Fern 0.1 | Mirendil | 53.2% |  | [Mirendil, "Introducing Fern"](https://mirendil.com/news/introducing-fern/) (own drop) |
| 2 | Nanbeige4.3-33B | Nanbeige | 50.8% | max | [Nanbeige4.3-33B model card](https://huggingface.co/Nanbeige/Nanbeige4.3-33B) (own drop) |
| 3 | Qwen3.8 | Alibaba | 41.9% | max | [Nanbeige4.3-33B model card](https://huggingface.co/Nanbeige/Nanbeige4.3-33B) |
| 4 | Muse Glimmer-30B | Meta | 17.7% | max | [Nanbeige4.3-33B model card](https://huggingface.co/Nanbeige/Nanbeige4.3-33B) |
| 5 | Nemotron 3 Ultra | NVIDIA | 16.1% |  | [Mirendil, "Introducing Fern"](https://mirendil.com/news/introducing-fern/) |

1 more: https://handyai.news/api/benchmarks/swe-atlas-qna

## TAU-bench Airline

6 models from 2 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Claude Sonnet 4 | Anthropic | 60% |  | [Claude Opus 4.1](https://handyai.news/modeldrop/claude-opus-41) |
| 2 | Claude Opus 4 | Anthropic | 59.6% |  | [Claude Opus 4.1](https://handyai.news/modeldrop/claude-opus-41) |
| 3 | Claude 3.7 Sonnet | Anthropic | 58.4% |  | [Claude Opus 4 & Sonnet 4](https://handyai.news/modeldrop/claude-4) |
| 4 | Claude Opus 4.1 | Anthropic | 56% |  | [Claude Opus 4.1](https://handyai.news/modeldrop/claude-opus-41) (own drop) |
| 5 | o3 | OpenAI | 52% |  | [Claude Opus 4.1](https://handyai.news/modeldrop/claude-opus-41) |

1 more: https://handyai.news/api/benchmarks/tau-bench-airline

## Artificial Analysis Coding Agent Index

5 models from 2 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Claude Opus 5.5 | Anthropic | 62 |  | [Gemini 4 Carbon](https://handyai.news/modeldrop/gemini-4-carbon) |
| 2 | Gemini 4 Carbon | Google DeepMind | 59 |  | [Gemini 4 Carbon](https://handyai.news/modeldrop/gemini-4-carbon) (own drop) |
| 3 | Grok 4.7 | SpaceXAI | 56 |  | [Grok 4.7](https://handyai.news/modeldrop/grok-47) (own drop) |
| 4 | Gemini 4 Argon | Google DeepMind | 53 |  | [Gemini 4 Carbon](https://handyai.news/modeldrop/gemini-4-carbon) |
| 5 | Grok 4.6 | SpaceXAI | 47 |  | [Grok 4.7](https://handyai.news/modeldrop/grok-47) |

## EnterpriseOps-Gym

5 models from 2 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Nanbeige4.3-33B | Nanbeige | 41% | max | [Nanbeige4.3-33B model card](https://huggingface.co/Nanbeige/Nanbeige4.3-33B) (own drop) |
| 2 | Qwen3.8 | Alibaba | 40.2% | max | [Nanbeige4.3-33B model card](https://huggingface.co/Nanbeige/Nanbeige4.3-33B) |
| 3 | Muse Glimmer-30B | Meta | 30.5% | max | [Nanbeige4.3-33B model card](https://huggingface.co/Nanbeige/Nanbeige4.3-33B) |
| 4 | LFM3-1.7B | Liquid AI | 28.2% |  | [Liquid AI, "Introducing LFM3"](https://www.liquid.ai/blog/lfm3) (own drop) |
| 5 | Gemma 4 | Google DeepMind | 27.3% | max | [Nanbeige4.3-33B model card](https://huggingface.co/Nanbeige/Nanbeige4.3-33B) |

## FrontierMath

5 models from 2 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | GPT-5.1 | OpenAI | 26.7% | high | [GPT-5.1](https://handyai.news/modeldrop/gpt-51) (own drop) |
| 2 | GPT-5 | OpenAI | 26.3% | high | [GPT-5.1](https://handyai.news/modeldrop/gpt-51) |
| 3 | o3-mini | OpenAI | 9.2% | high | [o3-mini](https://handyai.news/modeldrop/o3-mini) (own drop) |
| 4 | o1-mini | OpenAI | 5.8% |  | [o3-mini](https://handyai.news/modeldrop/o3-mini) |
| 5 | o1 | OpenAI | 5.5% |  | [o3-mini](https://handyai.news/modeldrop/o3-mini) |

## FrontierSWE v2

Frontier software engineering tasks. 5 models from 2 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | GPT-6 Astra | OpenAI | 65.5% | max | [Gemini 4 Carbon](https://handyai.news/modeldrop/gemini-4-carbon) |
| 2 | Claude Opus 5.5 | Anthropic | 62.3% | max | [Gemini 4 Carbon](https://handyai.news/modeldrop/gemini-4-carbon) |
| 3 | Gemini 4 Carbon | Google DeepMind | 61.2% | max | [Gemini 4 Carbon](https://handyai.news/modeldrop/gemini-4-carbon) (own drop) |
| 4 | Fable 5.1 | Anthropic | 56.3% | max | [Gemini 4 Carbon](https://handyai.news/modeldrop/gemini-4-carbon) |
| 5 | Gemini 4 Argon | Google DeepMind | 55% | max | [Gemini 4 Carbon](https://handyai.news/modeldrop/gemini-4-carbon) |

## Professional Capture-the-Flag Challenges

5 models from 2 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | GPT-5.2-Codex | OpenAI | 88.1% |  | [GPT-5.2-Codex](https://handyai.news/modeldrop/gpt-52-codex) (own drop) |
| 2 | GPT-5.2 | OpenAI | 82.4% |  | [GPT-5.2-Codex](https://handyai.news/modeldrop/gpt-52-codex) |
| 3 | GPT-5.1-Codex-Max | OpenAI | 75.6% |  | [GPT-5.2-Codex](https://handyai.news/modeldrop/gpt-52-codex) |
| 4 | GPT-5-Codex | OpenAI | 50.5% |  | [GPT-5.2-Codex](https://handyai.news/modeldrop/gpt-52-codex) |
| 5 | GPT-5.1 | OpenAI | 43% | xhigh | [GPT-5.1-Codex-Max](https://handyai.news/modeldrop/gpt-51-codex-max) |

## RULER @ 1M

5 models from 2 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Nemotron 3 Ultra | NVIDIA | 94.7% |  | [Nemotron 3 Ultra](https://handyai.news/modeldrop/nemotron-3-ultra) (own drop) |
| 2 | Nemotron 3 Super | NVIDIA | 91.8% |  | [Nemotron 3 Super](https://handyai.news/modeldrop/nemotron-3-super) (own drop) |
| 3 | Qwen3.5-122B-A10B | Alibaba | 91.3% |  | [Nemotron 3 Super](https://handyai.news/modeldrop/nemotron-3-super) |
| 4 | Qwen3.5 397B | Alibaba | 90.1% |  | [Nemotron 3 Ultra](https://handyai.news/modeldrop/nemotron-3-ultra) |
| 5 | gpt-oss-120b | OpenAI | 22.3% |  | [Nemotron 3 Super](https://handyai.news/modeldrop/nemotron-3-super) |

## SWE-Lancer IC Diamond

5 models from 2 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | GPT-5.3-Codex | OpenAI | 81.4% |  | [GPT-5.3-Codex](https://handyai.news/modeldrop/gpt-53-codex) (own drop) |
| 2 | GPT-5.1-Codex-Max | OpenAI | 79.9% | xhigh | [GPT-5.1-Codex-Max](https://handyai.news/modeldrop/gpt-51-codex-max) (own drop) |
| 3 | GPT-5.2-Codex | OpenAI | 76% |  | [GPT-5.3-Codex](https://handyai.news/modeldrop/gpt-53-codex) |
| 4 | GPT-5.2 | OpenAI | 74.6% | high | [GPT-5.3-Codex](https://handyai.news/modeldrop/gpt-53-codex) |
| 5 | GPT-5.1-Codex | OpenAI | 66.3% | high | [GPT-5.1-Codex-Max](https://handyai.news/modeldrop/gpt-51-codex-max) |

## MRCR (1M)

4 models from 2 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Gemini 1.5 Pro | Google DeepMind | 82.6% |  | [Gemini 2.0 Flash](https://handyai.news/modeldrop/gemini-20-flash) |
| 2 | Gemini 1.5 Flash | Google DeepMind | 71.9% |  | [Gemini 2.0 Flash](https://handyai.news/modeldrop/gemini-20-flash) |
| 3 | Gemini 2.5 Flash | Google DeepMind | 66.3% |  | [Gemini 2.5 Flash](https://handyai.news/modeldrop/gemini-25-flash) (own drop) |
| 4 | Gemini 2.0 Flash | Google DeepMind | 48.2% |  | [Gemini 2.5 Flash](https://handyai.news/modeldrop/gemini-25-flash) |

## PaperBench-10

4 models from 2 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | GPT-5.2-Codex | OpenAI | 43% |  | [GPT-5.2-Codex](https://handyai.news/modeldrop/gpt-52-codex) (own drop) |
| 2 | GPT-5.1-Codex-Max | OpenAI | 40% | xhigh | [GPT-5.2-Codex](https://handyai.news/modeldrop/gpt-52-codex) |
| 3 | GPT-5.2 | OpenAI | 39% |  | [GPT-5.2-Codex](https://handyai.news/modeldrop/gpt-52-codex) |
| 4 | GPT-5.1 | OpenAI | 34% | xhigh | [GPT-5.1-Codex-Max](https://handyai.news/modeldrop/gpt-51-codex-max) |

## CyBench

3 models from 2 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Grok 4 | SpaceXAI | 43% |  | [Grok 4.1](https://handyai.news/modeldrop/grok-41) |
| 2 | Grok 4.1 | SpaceXAI | 39% |  | [Grok 4.1](https://handyai.news/modeldrop/grok-41) (own drop) |
| 3 | Grok Code Fast 1 | SpaceXAI | 22.5% |  | [Grok Code Fast 1](https://handyai.news/modeldrop/grok-code-fast-1) (own drop) |

## LiveBench Coding

3 models from 2 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | o3-mini | OpenAI | 84.6% | high | [o3-mini](https://handyai.news/modeldrop/o3-mini) (own drop) |
| 2 | o1 | OpenAI | 67.4% |  | [o3-mini](https://handyai.news/modeldrop/o3-mini) |
| 3 | o1-preview | OpenAI | 52.3% |  | [o1](https://handyai.news/modeldrop/o1) |

## MASK dishonesty rate

Lower is better. 3 models from 2 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Grok 4 | SpaceXAI | 43% |  | [Grok 4.1](https://handyai.news/modeldrop/grok-41) |
| 2 | Grok 4.1 | SpaceXAI | 49% |  | [Grok 4.1](https://handyai.news/modeldrop/grok-41) (own drop) |
| 3 | Grok Code Fast 1 | SpaceXAI | 71.9% |  | [Grok Code Fast 1](https://handyai.news/modeldrop/grok-code-fast-1) (own drop) |

## FrontierCode 2.0

Real coding tasks in open-source repos (Cognition), run by Artificial Analysis. 27 models from 1 drop.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Claude Opus 5.5 | Anthropic | 46.4% | max | [Artificial Analysis (2026-11-04)](https://artificialanalysis.ai/evaluations) |
| 2 | GPT-6 Astra | OpenAI | 42.4% | max | [Artificial Analysis (2026-11-04)](https://artificialanalysis.ai/evaluations) |
| 3 | GPT-6.1 Terra | OpenAI | 42.3% | max | [Artificial Analysis (2026-11-04)](https://artificialanalysis.ai/evaluations) |
| 4 | Claude Sonnet 5.5 | Anthropic | 41.7% | max | [Artificial Analysis (2026-11-04)](https://artificialanalysis.ai/evaluations) |
| 5 | Gemini 4 Carbon | Google DeepMind | 40.6% | high | [Artificial Analysis (2026-11-04)](https://artificialanalysis.ai/evaluations) |

22 more: https://handyai.news/api/benchmarks/frontiercode-2-0

## Artificial Analysis Intelligence Index v5.0

26 models from 1 drop.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Gemini 4 Carbon | Google DeepMind | 60 | high | [Artificial Analysis (2026-11-04)](https://artificialanalysis.ai/evaluations) |
| 2 | Claude Opus 5.5 | Anthropic | 59.1 | max | [Artificial Analysis (2026-11-04)](https://artificialanalysis.ai/evaluations) |
| 3 | GPT-6 Astra | OpenAI | 56.9 | max | [Artificial Analysis (2026-11-04)](https://artificialanalysis.ai/evaluations) |
| 4 | GPT-6.1 Sol | OpenAI | 54.6 | max | [Artificial Analysis (2026-11-04)](https://artificialanalysis.ai/evaluations) |
| 5 | Claude Sonnet 5.5 | Anthropic | 54.5 | max | [Artificial Analysis (2026-11-04)](https://artificialanalysis.ai/evaluations) |

21 more: https://handyai.news/api/benchmarks/artificial-analysis-intelligence-index-v5-0

## Vals Index v2.2

24 models from 1 drop.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Claude Sonnet 5.5 | Anthropic | 70.9% |  | [Vals Index v2.2 (2026-11-04)](https://www.vals.ai/benchmarks/vals_index) |
| 2 | Gemini 4 Carbon | Google DeepMind | 70.5% |  | [Vals Index v2.2 (2026-11-04)](https://www.vals.ai/benchmarks/vals_index) |
| 3 | Claude Opus 5.5 | Anthropic | 70.2% |  | [Vals Index v2.2 (2026-11-04)](https://www.vals.ai/benchmarks/vals_index) |
| 4 | Gemini 4 Argon | Google DeepMind | 69.2% |  | [Vals Index v2.2 (2026-11-04)](https://www.vals.ai/benchmarks/vals_index) |
| 5 | Muse Wonder Full | Meta | 66.8% |  | [Vals Index v2.2 (2026-11-04)](https://www.vals.ai/benchmarks/vals_index) |

19 more: https://handyai.news/api/benchmarks/vals-index-v2-2

## Artificial Analysis Cyber Index

14 models from 1 drop.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Kimi K3.1 Pro | Moonshot AI | 64.1 |  | [Artificial Analysis (2026-11-04)](https://artificialanalysis.ai/evaluations) |
| 2 | DeepSeek V4.1 Flash 1017 | DeepSeek | 56.8 | max | [Artificial Analysis (2026-11-04)](https://artificialanalysis.ai/evaluations) |
| 3 | Grok 4.7 | SpaceXAI | 56.4 | xhigh | [Artificial Analysis (2026-11-04)](https://artificialanalysis.ai/evaluations) |
| 4 | MiMo-V2.6-Pro | Xiaomi | 56.1 |  | [Artificial Analysis (2026-11-04)](https://artificialanalysis.ai/evaluations) |
| 5 | GPT-6 Luna | OpenAI | 52.7 | max | [Artificial Analysis (2026-11-04)](https://artificialanalysis.ai/evaluations) |

9 more: https://handyai.news/api/benchmarks/artificial-analysis-cyber-index

## Apex Bench v1.1

6 models from 1 drop.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | DeepSeek V4.1 Flash 1017 | DeepSeek | 72.1% |  | [DeepSeek V4.1 Flash 1017](https://handyai.news/modeldrop/deepseek-v41-flash-1017) (own drop) |
| 2 | Claude Opus 5 | Anthropic | 65.8% |  | [DeepSeek V4.1 Flash 1017](https://handyai.news/modeldrop/deepseek-v41-flash-1017) |
| 3 | DeepSeek V4.1 Flash | DeepSeek | 64.6% |  | [DeepSeek V4.1 Flash 1017](https://handyai.news/modeldrop/deepseek-v41-flash-1017) |
| 4 | GLM-5.3 | Z.ai | 56.6% |  | [DeepSeek V4.1 Flash 1017](https://handyai.news/modeldrop/deepseek-v41-flash-1017) |
| 5 | GPT-5.6 Sol | OpenAI | 54.3% |  | [DeepSeek V4.1 Flash 1017](https://handyai.news/modeldrop/deepseek-v41-flash-1017) |

1 more: https://handyai.news/api/benchmarks/apex-bench-v1-1

## BankerToolBench

6 models from 1 drop.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | MiniMax M3.1 Flash | MiniMax | 76.2 |  | [MiniMax M3.1 Flash](https://handyai.news/modeldrop/minimax-m31-flash) (own drop) |
| 2 | Claude Opus 5 | Anthropic | 76.2 |  | [MiniMax M3.1 Flash](https://handyai.news/modeldrop/minimax-m31-flash) |
| 3 | GPT-5.6 Sol | OpenAI | 74.6 |  | [MiniMax M3.1 Flash](https://handyai.news/modeldrop/minimax-m31-flash) |
| 4 | GLM-5.3 | Z.ai | 74.5 |  | [MiniMax M3.1 Flash](https://handyai.news/modeldrop/minimax-m31-flash) |
| 5 | DeepSeek V4.1 Flash | DeepSeek | 73.7 |  | [MiniMax M3.1 Flash](https://handyai.news/modeldrop/minimax-m31-flash) |

1 more: https://handyai.news/api/benchmarks/bankertoolbench

## Frontier Finance

6 models from 1 drop.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Fable 5.1 | Anthropic | 60.8% | max | [Kimi K3.1 Pro](https://handyai.news/modeldrop/kimi-k31-pro) |
| 2 | Kimi K3.1 Pro | Moonshot AI | 58.6% | max | [Kimi K3.1 Pro](https://handyai.news/modeldrop/kimi-k31-pro) (own drop) |
| 3 | GLM-5.3 | Z.ai | 57.5% | max | [Kimi K3.1 Pro](https://handyai.news/modeldrop/kimi-k31-pro) |
| 4 | Claude Opus 5.5 | Anthropic | 56.7% | max | [Kimi K3.1 Pro](https://handyai.news/modeldrop/kimi-k31-pro) |
| 5 | DeepSeek V4.1 Flash 1017 | DeepSeek | 54.5% | max | [Kimi K3.1 Pro](https://handyai.news/modeldrop/kimi-k31-pro) |

1 more: https://handyai.news/api/benchmarks/frontier-finance

## OfficeQA Pro v2

6 models from 1 drop.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Kimi K3.1 Pro | Moonshot AI | 64.4% | max | [Kimi K3.1 Pro](https://handyai.news/modeldrop/kimi-k31-pro) (own drop) |
| 2 | Claude Opus 5.5 | Anthropic | 62.2% | max | [Kimi K3.1 Pro](https://handyai.news/modeldrop/kimi-k31-pro) |
| 3 | Fable 5.1 | Anthropic | 54.5% | max | [Kimi K3.1 Pro](https://handyai.news/modeldrop/kimi-k31-pro) |
| 4 | GPT-6 Astra | OpenAI | 54.4% | max | [Kimi K3.1 Pro](https://handyai.news/modeldrop/kimi-k31-pro) |
| 5 | DeepSeek V4.1 Flash 1017 | DeepSeek | 40.4% | max | [Kimi K3.1 Pro](https://handyai.news/modeldrop/kimi-k31-pro) |

1 more: https://handyai.news/api/benchmarks/officeqa-pro-v2

## Tokens per second

6 models from 1 drop.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Grok Code Fast 1 | SpaceXAI | 160 |  | [Grok Code Fast 1](https://handyai.news/modeldrop/grok-code-fast-1) (own drop) |
| 2 | Gemini 2.5 Pro | Google DeepMind | 92.4 |  | [Grok Code Fast 1](https://handyai.news/modeldrop/grok-code-fast-1) |
| 3 | Qwen3-Coder | Alibaba | 80 |  | [Grok Code Fast 1](https://handyai.news/modeldrop/grok-code-fast-1) |
| 4 | Claude Sonnet 4 | Anthropic | 78.7 |  | [Grok Code Fast 1](https://handyai.news/modeldrop/grok-code-fast-1) |
| 5 | GPT-5 | OpenAI | 50.1 |  | [Grok Code Fast 1](https://handyai.news/modeldrop/grok-code-fast-1) |

1 more: https://handyai.news/api/benchmarks/tokens-per-second

## Artificial Analysis Coding Agent Index v2.0

5 models from 1 drop.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Claude Opus 5.5 | Anthropic | 62 | max | [GPT-6.1 Terra](https://handyai.news/modeldrop/gpt-61-terra) |
| 2 | Claude Sonnet 5.5 | Anthropic | 59 | max | [GPT-6.1 Terra](https://handyai.news/modeldrop/gpt-61-terra) |
| 3 | GPT-6.1 Sol | OpenAI | 54 | max | [GPT-6.1 Terra](https://handyai.news/modeldrop/gpt-61-terra) |
| 4 | GPT-6.1 Terra | OpenAI | 48 | max | [GPT-6.1 Terra](https://handyai.news/modeldrop/gpt-61-terra) (own drop) |
| 5 | GPT-6 Luna | OpenAI | 31 | max | [GPT-6.1 Terra](https://handyai.news/modeldrop/gpt-61-terra) |

## BenchCAD (with tools)

5 models from 1 drop.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Claude Opus 5.5 | Anthropic | 96.2% | max | [Kimi K3.1 Pro](https://handyai.news/modeldrop/kimi-k31-pro) |
| 2 | GPT-6 Astra | OpenAI | 95.9% | max | [Kimi K3.1 Pro](https://handyai.news/modeldrop/kimi-k31-pro) |
| 3 | Kimi K3.1 Pro | Moonshot AI | 92.7% | max | [Kimi K3.1 Pro](https://handyai.news/modeldrop/kimi-k31-pro) (own drop) |
| 4 | Fable 5.1 | Anthropic | 92.6% | max | [Kimi K3.1 Pro](https://handyai.news/modeldrop/kimi-k31-pro) |
| 5 | DeepSeek V4.1 Flash 1017 | DeepSeek | 81.6% | max | [Kimi K3.1 Pro](https://handyai.news/modeldrop/kimi-k31-pro) |

## CyberGym-E2E

5 models from 1 drop.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Mistral Large 4 | Mistral AI | 82% |  | [Mistral Large 4](https://handyai.news/modeldrop/mistral-large-4) (own drop) |
| 2 | MiMo-V2.6-Pro | Xiaomi | 79% |  | [Mistral Large 4](https://handyai.news/modeldrop/mistral-large-4) |
| 3 | Grok 4.7 | SpaceXAI | 74% |  | [Mistral Large 4](https://handyai.news/modeldrop/mistral-large-4) |
| 4 | Kimi K3 | Moonshot AI | 58% |  | [Mistral Large 4](https://handyai.news/modeldrop/mistral-large-4) |
| 5 | GLM-5.3 | Z.ai | 29% |  | [Mistral Large 4](https://handyai.news/modeldrop/mistral-large-4) |

## EQ-Bench3

5 models from 1 drop.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Grok 4.1 | SpaceXAI | 1586 |  | [Grok 4.1](https://handyai.news/modeldrop/grok-41) (own drop) |
| 2 | Kimi K2 | Moonshot AI | 1561 |  | [Grok 4.1](https://handyai.news/modeldrop/grok-41) |
| 3 | Gemini 2.5 Pro | Google DeepMind | 1460 |  | [Grok 4.1](https://handyai.news/modeldrop/grok-41) |
| 4 | Claude Opus 4 | Anthropic | 1304 |  | [Grok 4.1](https://handyai.news/modeldrop/grok-41) |
| 5 | Grok 4 | SpaceXAI | 1206 |  | [Grok 4.1](https://handyai.news/modeldrop/grok-41) |

## Factual error rate

Lower is better. 5 models from 1 drop.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | GPT-6 Astra | OpenAI | 4.2% | max | [GPT-6.1 Terra](https://handyai.news/modeldrop/gpt-61-terra) |
| 2 | GPT-6.1 Terra | OpenAI | 4.3% | max | [GPT-6.1 Terra](https://handyai.news/modeldrop/gpt-61-terra) (own drop) |
| 3 | GPT-6.1 Sol | OpenAI | 4.4% | max | [GPT-6.1 Terra](https://handyai.news/modeldrop/gpt-61-terra) |
| 4 | GPT-6 Luna | OpenAI | 18.4% | max | [GPT-6.1 Terra](https://handyai.news/modeldrop/gpt-61-terra) |
| 5 | GPT-5.6 Terra | OpenAI | 22.3% | max | [GPT-6.1 Terra](https://handyai.news/modeldrop/gpt-61-terra) |

## IFBench (loose)

5 models from 1 drop.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Gemma 4 26B A4B | Google DeepMind | 77.3% |  | [Nemotron 3.5 Lightning](https://handyai.news/modeldrop/nemotron-35-lightning) |
| 2 | Nemotron 3 Nano | NVIDIA | 72.2% |  | [Nemotron 3.5 Lightning](https://handyai.news/modeldrop/nemotron-35-lightning) |
| 3 | Nemotron 3.5 Lightning | NVIDIA | 71.9% |  | [Nemotron 3.5 Lightning](https://handyai.news/modeldrop/nemotron-35-lightning) (own drop) |
| 4 | gpt-oss-20b | OpenAI | 68.5% |  | [Nemotron 3.5 Lightning](https://handyai.news/modeldrop/nemotron-35-lightning) |
| 5 | Qwen3.6 35B A3B | Alibaba | 63.7% |  | [Nemotron 3.5 Lightning](https://handyai.news/modeldrop/nemotron-35-lightning) |

## MemoryBench

5 models from 1 drop.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Muse Wonder | Meta | 85.6% | max | [Muse Wonder](https://handyai.news/modeldrop/muse-wonder) (own drop) |
| 2 | Muse Wonder Full | Meta | 85.5% | max | [Muse Wonder](https://handyai.news/modeldrop/muse-wonder) |
| 3 | Claude Opus 5.5 | Anthropic | 81.5% | max | [Muse Wonder](https://handyai.news/modeldrop/muse-wonder) |
| 4 | GPT-6 Astra | OpenAI | 81.4% | max | [Muse Wonder](https://handyai.news/modeldrop/muse-wonder) |
| 5 | Fable 5.1 | Anthropic | 79% | max | [Muse Wonder](https://handyai.news/modeldrop/muse-wonder) |

## MMMU-Pro Vision

5 models from 1 drop.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | GPT-6 Astra | OpenAI | 86.7% | max | [GPT-6.1 Terra](https://handyai.news/modeldrop/gpt-61-terra) |
| 2 | GPT-6.1 Sol | OpenAI | 86.3% | max | [GPT-6.1 Terra](https://handyai.news/modeldrop/gpt-61-terra) |
| 3 | GPT-6.1 Terra | OpenAI | 85.5% | max | [GPT-6.1 Terra](https://handyai.news/modeldrop/gpt-61-terra) (own drop) |
| 4 | GPT-5.6 Terra | OpenAI | 81.7% | max | [GPT-6.1 Terra](https://handyai.news/modeldrop/gpt-61-terra) |
| 5 | GPT-6 Luna | OpenAI | 79.9% | max | [GPT-6.1 Terra](https://handyai.news/modeldrop/gpt-61-terra) |

## Personal Superintelligence Index

5 models from 1 drop.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Muse Wonder Full | Meta | 59.1 | max | [Muse Wonder](https://handyai.news/modeldrop/muse-wonder) |
| 2 | Claude Opus 5.5 | Anthropic | 55.2 | max | [Muse Wonder](https://handyai.news/modeldrop/muse-wonder) |
| 3 | GPT-6 Astra | OpenAI | 54.5 | max | [Muse Wonder](https://handyai.news/modeldrop/muse-wonder) |
| 4 | Muse Wonder | Meta | 54.3 | max | [Muse Wonder](https://handyai.news/modeldrop/muse-wonder) (own drop) |
| 5 | Fable 5.1 | Anthropic | 49.9 | max | [Muse Wonder](https://handyai.news/modeldrop/muse-wonder) |

## SWE-sweep

5 models from 1 drop.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Gemini 4 Carbon | Google DeepMind | 9.6% | max | [Gemini 4 Carbon](https://handyai.news/modeldrop/gemini-4-carbon) (own drop) |
| 2 | GPT-6 Astra | OpenAI | 9.5% | max | [Gemini 4 Carbon](https://handyai.news/modeldrop/gemini-4-carbon) |
| 3 | Gemini 4 Argon | Google DeepMind | 4.3% | max | [Gemini 4 Carbon](https://handyai.news/modeldrop/gemini-4-carbon) |
| 4 | Claude Opus 5.5 | Anthropic | 3.6% | max | [Gemini 4 Carbon](https://handyai.news/modeldrop/gemini-4-carbon) |
| 5 | Fable 5.1 | Anthropic | 2.9% | max | [Gemini 4 Carbon](https://handyai.news/modeldrop/gemini-4-carbon) |

## Vals Finance Agent v2

5 models from 1 drop.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Gemini 3.8 Flash | Google DeepMind | 61.4% | max | [Gemini 3.8 Flash](https://handyai.news/modeldrop/gemini-38-flash) (own drop) |
| 2 | Claude Opus 5 | Anthropic | 58.6% | max | [Gemini 3.8 Flash](https://handyai.news/modeldrop/gemini-38-flash) |
| 3 | GPT-5.6 Terra | OpenAI | 54.4% | max | [Gemini 3.8 Flash](https://handyai.news/modeldrop/gemini-38-flash) |
| 4 | Claude Sonnet 5 | Anthropic | 53.9% | max | [Gemini 3.8 Flash](https://handyai.news/modeldrop/gemini-38-flash) |
| 5 | GPT-5.6 Sol | OpenAI | 53.8% | max | [Gemini 3.8 Flash](https://handyai.news/modeldrop/gemini-38-flash) |

## τ²-Bench

5 models from 1 drop.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Claude Sonnet 4.5 | Anthropic | 84.7% |  | [MiniMax-M2](https://handyai.news/modeldrop/minimax-m2) |
| 2 | GPT-5 | OpenAI | 80.1% |  | [MiniMax-M2](https://handyai.news/modeldrop/minimax-m2) |
| 3 | MiniMax-M2 | MiniMax | 77.2% |  | [MiniMax-M2](https://handyai.news/modeldrop/minimax-m2) (own drop) |
| 4 | GLM-4.6 | Z.ai | 75.9% |  | [MiniMax-M2](https://handyai.news/modeldrop/minimax-m2) |
| 5 | Kimi K2 0905 | Moonshot AI | 70.3% |  | [MiniMax-M2](https://handyai.news/modeldrop/minimax-m2) |

## Agents' Last Exam v1.1

4 models from 1 drop.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Gemini 4 Carbon | Google DeepMind | 77.4% | max | [Gemini 4 Carbon](https://handyai.news/modeldrop/gemini-4-carbon) (own drop) |
| 2 | GPT-6 Astra | OpenAI | 73.1% | max | [Gemini 4 Carbon](https://handyai.news/modeldrop/gemini-4-carbon) |
| 3 | Claude Opus 5.5 | Anthropic | 71.6% | max | [Gemini 4 Carbon](https://handyai.news/modeldrop/gemini-4-carbon) |
| 4 | Gemini 4 Argon | Google DeepMind | 69% | max | [Gemini 4 Carbon](https://handyai.news/modeldrop/gemini-4-carbon) |

## ArenaHard

4 models from 1 drop.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Mistral Medium 3 | Mistral AI | 97.1% |  | [Mistral Medium 3](https://handyai.news/modeldrop/mistral-medium-3) (own drop) |
| 2 | GPT-4o | OpenAI | 95.4% |  | [Mistral Medium 3](https://handyai.news/modeldrop/mistral-medium-3) |
| 3 | Claude 3.7 Sonnet | Anthropic | 93.2% |  | [Mistral Medium 3](https://handyai.news/modeldrop/mistral-medium-3) |
| 4 | Llama 4 Maverick | Meta | 91.8% |  | [Mistral Medium 3](https://handyai.news/modeldrop/mistral-medium-3) |

## Chartography

4 models from 1 drop.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Claude Sonnet 5.5 | Anthropic | 61.6% |  | [Claude Haiku 5.5](https://handyai.news/modeldrop/claude-haiku-55) |
| 2 | Claude Haiku 5.5 | Anthropic | 46.4% | max | [Claude Haiku 5.5](https://handyai.news/modeldrop/claude-haiku-55) (own drop) |
| 3 | GPT-6 Luna | OpenAI | 29.1% |  | [Claude Haiku 5.5](https://handyai.news/modeldrop/claude-haiku-55) |
| 4 | Claude Haiku 4.5 | Anthropic | 6.4% |  | [Claude Haiku 5.5](https://handyai.news/modeldrop/claude-haiku-55) |

## ChartQA

4 models from 1 drop.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Llama 4 Maverick | Meta | 90% |  | [Llama 4 Maverick & Scout](https://handyai.news/modeldrop/llama-4) (own drop) |
| 2 | Llama 4 Scout | Meta | 88.8% |  | [Llama 4 Maverick & Scout](https://handyai.news/modeldrop/llama-4) (own drop) |
| 3 | Gemini 2.0 Flash | Google DeepMind | 88.3% |  | [Llama 4 Maverick & Scout](https://handyai.news/modeldrop/llama-4) |
| 4 | GPT-4o | OpenAI | 85.7% |  | [Llama 4 Maverick & Scout](https://handyai.news/modeldrop/llama-4) |

## Claw-Eval pass^3

4 models from 1 drop.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Nanbeige4.3-33B | Nanbeige | 78.3% | max | [Nanbeige4.3-33B model card](https://huggingface.co/Nanbeige/Nanbeige4.3-33B) (own drop) |
| 2 | Qwen3.8 | Alibaba | 75.2% | max | [Nanbeige4.3-33B model card](https://huggingface.co/Nanbeige/Nanbeige4.3-33B) |
| 3 | Muse Glimmer-30B | Meta | 57.3% | max | [Nanbeige4.3-33B model card](https://huggingface.co/Nanbeige/Nanbeige4.3-33B) |
| 4 | Gemma 4 | Google DeepMind | 44.2% | max | [Nanbeige4.3-33B model card](https://huggingface.co/Nanbeige/Nanbeige4.3-33B) |

## CursorBench 3.2

4 models from 1 drop.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Fable 5 | Anthropic | 70.5% | max | [Grok 4.6](https://handyai.news/modeldrop/grok-46) |
| 2 | Grok 4.6 | SpaceXAI | 69.9% | high | [Grok 4.6](https://handyai.news/modeldrop/grok-46) (own drop) |
| 3 | GPT-5.6 Sol | OpenAI | 67.2% | max | [Grok 4.6](https://handyai.news/modeldrop/grok-46) |
| 4 | Grok 4.5 | SpaceXAI | 66.7% | high | [Grok 4.6](https://handyai.news/modeldrop/grok-46) |

## DeepPlanning

4 models from 1 drop.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Nanbeige4.3-33B | Nanbeige | 63.3% | max | [Nanbeige4.3-33B model card](https://huggingface.co/Nanbeige/Nanbeige4.3-33B) (own drop) |
| 2 | Qwen3.8 | Alibaba | 49.2% | max | [Nanbeige4.3-33B model card](https://huggingface.co/Nanbeige/Nanbeige4.3-33B) |
| 3 | Muse Glimmer-30B | Meta | 21.4% | max | [Nanbeige4.3-33B model card](https://huggingface.co/Nanbeige/Nanbeige4.3-33B) |
| 4 | Gemma 4 | Google DeepMind | 19.4% | max | [Nanbeige4.3-33B model card](https://huggingface.co/Nanbeige/Nanbeige4.3-33B) |

## DocVQA

4 models from 1 drop.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Mistral Medium 3 | Mistral AI | 95.3% |  | [Mistral Medium 3](https://handyai.news/modeldrop/mistral-medium-3) (own drop) |
| 2 | Llama 4 Maverick | Meta | 94.1% |  | [Mistral Medium 3](https://handyai.news/modeldrop/mistral-medium-3) |
| 3 | GPT-4o | OpenAI | 85.9% |  | [Mistral Medium 3](https://handyai.news/modeldrop/mistral-medium-3) |
| 4 | Claude 3.7 Sonnet | Anthropic | 84.3% |  | [Mistral Medium 3](https://handyai.news/modeldrop/mistral-medium-3) |

## EEBench

4 models from 1 drop.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Grok 4.7 | SpaceXAI | 64% | xhigh | [Grok 4.7](https://handyai.news/modeldrop/grok-47) (own drop) |
| 2 | Fable 5.1 | Anthropic | 56.4% | max | [Grok 4.7](https://handyai.news/modeldrop/grok-47) |
| 3 | Grok 4.6 | SpaceXAI | 53% | high | [Grok 4.7](https://handyai.news/modeldrop/grok-47) |
| 4 | GPT-5.6 Sol | OpenAI | 39.4% | max | [Grok 4.7](https://handyai.news/modeldrop/grok-47) |

## Finance Agent v2

4 models from 1 drop.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | GLM-5.3 | Z.ai | 55.8% |  | [Mistral Large 4](https://handyai.news/modeldrop/mistral-large-4) |
| 2 | Mistral Large 4 | Mistral AI | 54.7% |  | [Mistral Large 4](https://handyai.news/modeldrop/mistral-large-4) (own drop) |
| 3 | GPT-6 Astra | OpenAI | 53.5% |  | [Mistral Large 4](https://handyai.news/modeldrop/mistral-large-4) |
| 4 | Kimi K3 | Moonshot AI | 53.1% |  | [Mistral Large 4](https://handyai.news/modeldrop/mistral-large-4) |

## GDM-MRCR v2

4 models from 1 drop.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Gemini 3.6 Flash | Google DeepMind | 91.8% | max | [Gemini 3.6 Flash](https://handyai.news/modeldrop/gemini-36-flash) (own drop) |
| 2 | Grok 4.5 | SpaceXAI | 81.4% | max | [Gemini 3.6 Flash](https://handyai.news/modeldrop/gemini-36-flash) |
| 3 | GPT-5.6 Luna | OpenAI | 74.8% | max | [Gemini 3.6 Flash](https://handyai.news/modeldrop/gemini-36-flash) |
| 4 | Claude Sonnet 5 | Anthropic | 71.6% | max | [Gemini 3.6 Flash](https://handyai.news/modeldrop/gemini-36-flash) |

## GDPval rubrics

4 models from 1 drop.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Qwen3.8 | Alibaba | 86.4% | max | [Nanbeige4.3-33B model card](https://huggingface.co/Nanbeige/Nanbeige4.3-33B) |
| 2 | Nanbeige4.3-33B | Nanbeige | 84.8% | max | [Nanbeige4.3-33B model card](https://huggingface.co/Nanbeige/Nanbeige4.3-33B) (own drop) |
| 3 | Muse Glimmer-30B | Meta | 75.8% | max | [Nanbeige4.3-33B model card](https://huggingface.co/Nanbeige/Nanbeige4.3-33B) |
| 4 | Gemma 4 | Google DeepMind | 71.4% | max | [Nanbeige4.3-33B model card](https://huggingface.co/Nanbeige/Nanbeige4.3-33B) |

## HealthBench Hard

4 models from 1 drop.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Muse Spark | Meta | 42.8% |  | [Muse Spark](https://handyai.news/modeldrop/muse-spark) (own drop) |
| 2 | GPT-5.4 | OpenAI | 40.1% | xhigh | [Muse Spark](https://handyai.news/modeldrop/muse-spark) |
| 3 | Gemini 3.1 Pro | Google DeepMind | 20.6% | high | [Muse Spark](https://handyai.news/modeldrop/muse-spark) |
| 4 | Claude Opus 4.6 | Anthropic | 14.8% | max | [Muse Spark](https://handyai.news/modeldrop/muse-spark) |

## HLE-Diamond (no tools)

4 models from 1 drop.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Muse Spark Open | Meta | 26.1% | xhigh | [Muse Spark Open](https://handyai.news/modeldrop/muse-spark-open) (own drop) |
| 2 | Kimi K3 | Moonshot AI | 22.2% | max | [Muse Spark Open](https://handyai.news/modeldrop/muse-spark-open) |
| 3 | Qwen3.8-2.4T-A95B | Alibaba | 17.3% | xhigh | [Muse Spark Open](https://handyai.news/modeldrop/muse-spark-open) |
| 4 | GLM-5.3 | Z.ai | 16.4% | max | [Muse Spark Open](https://handyai.news/modeldrop/muse-spark-open) |

## HumanEval

4 models from 1 drop.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Mistral Medium 3 | Mistral AI | 92.1% |  | [Mistral Medium 3](https://handyai.news/modeldrop/mistral-medium-3) (own drop) |
| 2 | Claude 3.7 Sonnet | Anthropic | 92.1% |  | [Mistral Medium 3](https://handyai.news/modeldrop/mistral-medium-3) |
| 3 | GPT-4o | OpenAI | 91.5% |  | [Mistral Medium 3](https://handyai.news/modeldrop/mistral-medium-3) |
| 4 | Llama 4 Maverick | Meta | 85.4% |  | [Mistral Medium 3](https://handyai.news/modeldrop/mistral-medium-3) |

## MathArena Apex

4 models from 1 drop.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Gemini 3 Pro | Google DeepMind | 23.4% |  | [Gemini 3 Pro](https://handyai.news/modeldrop/gemini-3-pro) (own drop) |
| 2 | Claude Sonnet 4.5 | Anthropic | 1.6% |  | [Gemini 3 Pro](https://handyai.news/modeldrop/gemini-3-pro) |
| 3 | GPT-5.1 | OpenAI | 1% |  | [Gemini 3 Pro](https://handyai.news/modeldrop/gemini-3-pro) |
| 4 | Gemini 2.5 Pro | Google DeepMind | 0.5% |  | [Gemini 3 Pro](https://handyai.news/modeldrop/gemini-3-pro) |

## MathVista

4 models from 1 drop.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Llama 4 Maverick | Meta | 73.7% |  | [Llama 4 Maverick & Scout](https://handyai.news/modeldrop/llama-4) (own drop) |
| 2 | Gemini 2.0 Flash | Google DeepMind | 73.1% |  | [Llama 4 Maverick & Scout](https://handyai.news/modeldrop/llama-4) |
| 3 | Llama 4 Scout | Meta | 70.7% |  | [Llama 4 Maverick & Scout](https://handyai.news/modeldrop/llama-4) (own drop) |
| 4 | GPT-4o | OpenAI | 63.8% |  | [Llama 4 Maverick & Scout](https://handyai.news/modeldrop/llama-4) |

## Meta Internal Coding Bench

4 models from 1 drop.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Claude Opus 5 | Anthropic | 79.4% | max | [Muse Spark 1.2](https://handyai.news/modeldrop/muse-spark-12) |
| 2 | Muse Spark 1.2 | Meta | 70.6% | xhigh | [Muse Spark 1.2](https://handyai.news/modeldrop/muse-spark-12) (own drop) |
| 3 | Muse Spark 1.1 | Meta | 68.3% | xhigh | [Muse Spark 1.2](https://handyai.news/modeldrop/muse-spark-12) |
| 4 | GPT-5.6 Terra | OpenAI | 65.4% | max | [Muse Spark 1.2](https://handyai.news/modeldrop/muse-spark-12) |

## MLE-Bench

4 models from 1 drop.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Claude Sonnet 5 | Anthropic | 66.9% | max | [Gemini 3.6 Flash](https://handyai.news/modeldrop/gemini-36-flash) |
| 2 | Gemini 3.6 Flash | Google DeepMind | 63.9% | max | [Gemini 3.6 Flash](https://handyai.news/modeldrop/gemini-36-flash) (own drop) |
| 3 | GPT-5.6 Luna | OpenAI | 47.6% | max | [Gemini 3.6 Flash](https://handyai.news/modeldrop/gemini-36-flash) |
| 4 | Grok 4.5 | SpaceXAI | 43.2% | max | [Gemini 3.6 Flash](https://handyai.news/modeldrop/gemini-36-flash) |

## PaperBench

4 models from 1 drop.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Qwen3.8-Max | Alibaba | 93% |  | [Qwen3.8-Max](https://handyai.news/modeldrop/qwen38-max) (own drop) |
| 2 | GPT-5.6 Sol | OpenAI | 90.5% | max | [Qwen3.8-Max](https://handyai.news/modeldrop/qwen38-max) |
| 3 | Fable 5 | Anthropic | 88.8% | max | [Qwen3.8-Max](https://handyai.news/modeldrop/qwen38-max) |
| 4 | Claude Opus 4.8 | Anthropic | 80.3% | max | [Qwen3.8-Max](https://handyai.news/modeldrop/qwen38-max) |

## SEC-Bench Pro

4 models from 1 drop.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | GPT-5.6 Sol | OpenAI | 71.2% | max | [GPT-5.6 Sol, Terra, & Luna](https://handyai.news/modeldrop/gpt-56-sol-terra-and-luna) (own drop) |
| 2 | GPT-5.6 Terra | OpenAI | 57.7% | max | [GPT-5.6 Sol, Terra, & Luna](https://handyai.news/modeldrop/gpt-56-sol-terra-and-luna) |
| 3 | GPT-5.6 Luna | OpenAI | 48.9% | max | [GPT-5.6 Sol, Terra, & Luna](https://handyai.news/modeldrop/gpt-56-sol-terra-and-luna) |
| 4 | GPT-5.5 | OpenAI | 45.8% | xhigh | [GPT-5.6 Sol, Terra, & Luna](https://handyai.news/modeldrop/gpt-56-sol-terra-and-luna) |

## SimpleQA hallucination rate

Lower is better. 4 models from 1 drop.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | GPT-4.5 | OpenAI | 37.1% |  | [GPT-4.5](https://handyai.news/modeldrop/gpt-45) (own drop) |
| 2 | o1 | OpenAI | 44% |  | [GPT-4.5](https://handyai.news/modeldrop/gpt-45) |
| 3 | GPT-4o | OpenAI | 61.8% |  | [GPT-4.5](https://handyai.news/modeldrop/gpt-45) |
| 4 | o3-mini | OpenAI | 80.3% |  | [GPT-4.5](https://handyai.news/modeldrop/gpt-45) |

## SkillsBench

4 models from 1 drop.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Nanbeige4.3-33B | Nanbeige | 61.1% | max | [Nanbeige4.3-33B model card](https://huggingface.co/Nanbeige/Nanbeige4.3-33B) (own drop) |
| 2 | Qwen3.8 | Alibaba | 56.4% | max | [Nanbeige4.3-33B model card](https://huggingface.co/Nanbeige/Nanbeige4.3-33B) |
| 3 | Muse Glimmer-30B | Meta | 39.7% | max | [Nanbeige4.3-33B model card](https://huggingface.co/Nanbeige/Nanbeige4.3-33B) |
| 4 | Gemma 4 | Google DeepMind | 28.9% | max | [Nanbeige4.3-33B model card](https://huggingface.co/Nanbeige/Nanbeige4.3-33B) |

## SuperGPQA

4 models from 1 drop.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Qwen3-Max | Alibaba | 65.1% |  | [Qwen3-Max](https://handyai.news/modeldrop/qwen3-max) (own drop) |
| 2 | Qwen3-235B-A22B-Instruct-2507 | Alibaba | 62.6% |  | [Qwen3-Max](https://handyai.news/modeldrop/qwen3-max) |
| 3 | DeepSeek V3.1 | DeepSeek | 59.8% |  | [Qwen3-Max](https://handyai.news/modeldrop/qwen3-max) |
| 4 | Claude Opus 4 | Anthropic | 56.5% |  | [Qwen3-Max](https://handyai.news/modeldrop/qwen3-max) |

## USAMO 2025

4 models from 1 drop.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Gemini Deep Think | Google DeepMind | 49.4% |  | [Grok 4](https://handyai.news/modeldrop/grok-4) |
| 2 | Grok 4 | SpaceXAI | 37.5% |  | [Grok 4](https://handyai.news/modeldrop/grok-4) (own drop) |
| 3 | Gemini 2.5 Pro | Google DeepMind | 34.5% |  | [Grok 4](https://handyai.news/modeldrop/grok-4) |
| 4 | o3 | OpenAI | 21.7% |  | [Grok 4](https://handyai.news/modeldrop/grok-4) |

## τ²-Bench (Weighted)

4 models from 1 drop.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Qwen3-Max | Alibaba | 74.8% |  | [Qwen3-Max](https://handyai.news/modeldrop/qwen3-max) (own drop) |
| 2 | Claude Opus 4 | Anthropic | 67.7% |  | [Qwen3-Max](https://handyai.news/modeldrop/qwen3-max) |
| 3 | Qwen3-235B-A22B-Instruct-2507 | Alibaba | 52.9% |  | [Qwen3-Max](https://handyai.news/modeldrop/qwen3-max) |
| 4 | DeepSeek V3.1 | DeepSeek | 46.4% |  | [Qwen3-Max](https://handyai.news/modeldrop/qwen3-max) |

## AIME 2024 4/4 reliability

3 models from 1 drop.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | o3-pro | OpenAI | 90% |  | [o3-pro](https://handyai.news/modeldrop/o3-pro) (own drop) |
| 2 | o3 | OpenAI | 80% | medium | [o3-pro](https://handyai.news/modeldrop/o3-pro) |
| 3 | o1-pro | OpenAI | 80% |  | [o3-pro](https://handyai.news/modeldrop/o3-pro) |

## ARC-AGI-3

3 models from 1 drop.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | GPT-6 Astra | OpenAI | 99.9% |  | [GPT-6 Astra](https://handyai.news/modeldrop/gpt-6-astra) (own drop) |
| 2 | Claude Opus 5 | Anthropic | 30.2% |  | [GPT-6 Astra](https://handyai.news/modeldrop/gpt-6-astra) |
| 3 | GPT-5.6 Sol | OpenAI | 7.8% |  | [GPT-6 Astra](https://handyai.news/modeldrop/gpt-6-astra) |

## Arena AI text score

3 models from 1 drop.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Gemma 4 | Google DeepMind | 1452 |  | [Gemma 4](https://handyai.news/modeldrop/gemma-4) (own drop) |
| 2 | Gemma 4 26B A4B | Google DeepMind | 1441 |  | [Gemma 4](https://handyai.news/modeldrop/gemma-4) |
| 3 | Gemma 3 27B | Google DeepMind | 1365 |  | [Gemma 4](https://handyai.news/modeldrop/gemma-4) |

## Arena Hard

3 models from 1 drop.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | DeepSeek R1 | DeepSeek | 92% |  | [Llama 3.1 Nemotron Ultra](https://handyai.news/modeldrop/llama-31-nemotron-ultra) |
| 2 | Llama 3.1 Nemotron Ultra | NVIDIA | 87% |  | [Llama 3.1 Nemotron Ultra](https://handyai.news/modeldrop/llama-31-nemotron-ultra) (own drop) |
| 3 | Llama 3.1 405B | Meta | 66.2% |  | [Llama 3.1 Nemotron Ultra](https://handyai.news/modeldrop/llama-31-nemotron-ultra) |

## BioMysteryBench hard

3 models from 1 drop.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Mythos 5 | Anthropic | 46.1% |  | [Fable 5 / Mythos 5](https://handyai.news/modeldrop/fable-5-mythos-5) (own drop) |
| 2 | Claude Opus 4.8 | Anthropic | 40% |  | [Fable 5 / Mythos 5](https://handyai.news/modeldrop/fable-5-mythos-5) |
| 3 | Claude Mythos Preview | Anthropic | 29.6% |  | [Fable 5 / Mythos 5](https://handyai.news/modeldrop/fable-5-mythos-5) |

## BioMysteryBench human-solved

3 models from 1 drop.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Mythos 5 | Anthropic | 83.9% |  | [Fable 5 / Mythos 5](https://handyai.news/modeldrop/fable-5-mythos-5) (own drop) |
| 2 | Claude Mythos Preview | Anthropic | 82.6% |  | [Fable 5 / Mythos 5](https://handyai.news/modeldrop/fable-5-mythos-5) |
| 3 | Claude Opus 4.8 | Anthropic | 80.4% |  | [Fable 5 / Mythos 5](https://handyai.news/modeldrop/fable-5-mythos-5) |

## Codeforces 4/4 reliability

3 models from 1 drop.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | o3-pro | OpenAI | 2301 |  | [o3-pro](https://handyai.news/modeldrop/o3-pro) (own drop) |
| 2 | o3 | OpenAI | 2011 | medium | [o3-pro](https://handyai.news/modeldrop/o3-pro) |
| 3 | o1-pro | OpenAI | 1423 |  | [o3-pro](https://handyai.news/modeldrop/o3-pro) |

## Creative Writing v3

3 models from 1 drop.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Grok 4.1 | SpaceXAI | 1721.9 |  | [Grok 4.1](https://handyai.news/modeldrop/grok-41) (own drop) |
| 2 | o3 | OpenAI | 1696.4 |  | [Grok 4.1](https://handyai.news/modeldrop/grok-41) |
| 3 | Claude Sonnet 4.5 | Anthropic | 1648.7 |  | [Grok 4.1](https://handyai.news/modeldrop/grok-41) |

## CVE-Bench

3 models from 1 drop.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | GPT-5.2-Codex | OpenAI | 87% |  | [GPT-5.2-Codex](https://handyai.news/modeldrop/gpt-52-codex) (own drop) |
| 2 | GPT-5.1-Codex-Max | OpenAI | 80% |  | [GPT-5.2-Codex](https://handyai.news/modeldrop/gpt-52-codex) |
| 3 | GPT-5.2 | OpenAI | 69% |  | [GPT-5.2-Codex](https://handyai.news/modeldrop/gpt-52-codex) |

## Cyber Range

3 models from 1 drop.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | GPT-5.1-Codex-Max | OpenAI | 81.8% |  | [GPT-5.2-Codex](https://handyai.news/modeldrop/gpt-52-codex) |
| 2 | GPT-5.2-Codex | OpenAI | 72.7% |  | [GPT-5.2-Codex](https://handyai.news/modeldrop/gpt-52-codex) (own drop) |
| 3 | GPT-5.2 | OpenAI | 63.6% |  | [GPT-5.2-Codex](https://handyai.news/modeldrop/gpt-52-codex) |

## CyberGym Level 1 leaderboard

3 models from 1 drop.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Wiz Atlas |  | 90.9% |  | [MAI-Cyber-1-Flash](https://handyai.news/modeldrop/mai-cyber-1-flash) |
| 2 | MAI-Cyber-1-Flash | Microsoft | 86.3% |  | [MAI-Cyber-1-Flash](https://handyai.news/modeldrop/mai-cyber-1-flash) (own drop) |
| 3 | Claude Mythos Preview | Anthropic | 83.1% |  | [MAI-Cyber-1-Flash](https://handyai.news/modeldrop/mai-cyber-1-flash) |

## Cybersecurity Capture The Flag Challenges

3 models from 1 drop.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | GPT-5.3-Codex | OpenAI | 77.6% |  | [GPT-5.3-Codex](https://handyai.news/modeldrop/gpt-53-codex) (own drop) |
| 2 | GPT-5.2 | OpenAI | 67.7% | high | [GPT-5.3-Codex](https://handyai.news/modeldrop/gpt-53-codex) |
| 3 | GPT-5.2-Codex | OpenAI | 67.4% |  | [GPT-5.3-Codex](https://handyai.news/modeldrop/gpt-53-codex) |

## Dense200

3 models from 1 drop.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Mistral Large 4 | Mistral AI | 42% |  | [Mistral Large 4](https://handyai.news/modeldrop/mistral-large-4) (own drop) |
| 2 | GPT-6 Astra | OpenAI | 41.5% |  | [Mistral Large 4](https://handyai.news/modeldrop/mistral-large-4) |
| 3 | Kimi K3 | Moonshot AI | 28.9% |  | [Mistral Large 4](https://handyai.news/modeldrop/mistral-large-4) |

## FrontierMath Tier 4

3 models from 1 drop.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | GPT-6 Astra | OpenAI | 97.6% |  | [GPT-6 Astra](https://handyai.news/modeldrop/gpt-6-astra) (own drop) |
| 2 | Fable 5.1 | Anthropic | 87.8% |  | [GPT-6 Astra](https://handyai.news/modeldrop/gpt-6-astra) |
| 3 | GPT-5.6 Sol | OpenAI | 83% |  | [GPT-6 Astra](https://handyai.news/modeldrop/gpt-6-astra) |

## GPQA Diamond 4/4 reliability

3 models from 1 drop.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | o3-pro | OpenAI | 76% |  | [o3-pro](https://handyai.news/modeldrop/o3-pro) (own drop) |
| 2 | o1-pro | OpenAI | 74% |  | [o3-pro](https://handyai.news/modeldrop/o3-pro) |
| 3 | o3 | OpenAI | 67% | medium | [o3-pro](https://handyai.news/modeldrop/o3-pro) |

## HMMT Feb25

3 models from 1 drop.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Nemotron 3 Super | NVIDIA | 93.7% |  | [Nemotron 3 Super](https://handyai.news/modeldrop/nemotron-3-super) (own drop) |
| 2 | Qwen3.5-122B-A10B | Alibaba | 91.4% |  | [Nemotron 3 Super](https://handyai.news/modeldrop/nemotron-3-super) |
| 3 | gpt-oss-120b | OpenAI | 90% |  | [Nemotron 3 Super](https://handyai.news/modeldrop/nemotron-3-super) |

## MATH

3 models from 1 drop.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Gemini 2.0 Flash | Google DeepMind | 89.7% |  | [Gemini 2.0 Flash](https://handyai.news/modeldrop/gemini-20-flash) (own drop) |
| 2 | Gemini 1.5 Pro | Google DeepMind | 86.5% |  | [Gemini 2.0 Flash](https://handyai.news/modeldrop/gemini-20-flash) |
| 3 | Gemini 1.5 Flash | Google DeepMind | 77.9% |  | [Gemini 2.0 Flash](https://handyai.news/modeldrop/gemini-20-flash) |

## MLE-Bench-30

3 models from 1 drop.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | GPT-5.1-Codex-Max | OpenAI | 17% |  | [GPT-5.2-Codex](https://handyai.news/modeldrop/gpt-52-codex) |
| 2 | GPT-5.2 | OpenAI | 16% |  | [GPT-5.2-Codex](https://handyai.news/modeldrop/gpt-52-codex) |
| 3 | GPT-5.2-Codex | OpenAI | 10% |  | [GPT-5.2-Codex](https://handyai.news/modeldrop/gpt-52-codex) (own drop) |

## MRCR 512K-1M

3 models from 1 drop.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Muse Spark 1.3 | Meta | 98.1% | max | [Muse Spark 1.3](https://handyai.news/modeldrop/muse-spark-13) (own drop) |
| 2 | GPT-5.6 Sol | OpenAI | 73.8% | max | [Muse Spark 1.3](https://handyai.news/modeldrop/muse-spark-13) |
| 3 | Muse Spark 1.2 | Meta | 55.5% | xhigh | [Muse Spark 1.3](https://handyai.news/modeldrop/muse-spark-13) |

## MultiChallenge

3 models from 1 drop.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | GPT-4.5 | OpenAI | 43.8% |  | [GPT-4.1](https://handyai.news/modeldrop/gpt-41) |
| 2 | GPT-4.1 | OpenAI | 38.3% |  | [GPT-4.1](https://handyai.news/modeldrop/gpt-41) (own drop) |
| 3 | GPT-4o | OpenAI | 27.8% |  | [GPT-4.1](https://handyai.news/modeldrop/gpt-41) |

## Natural2Code

3 models from 1 drop.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Gemini 2.0 Flash | Google DeepMind | 92.9% |  | [Gemini 2.0 Flash](https://handyai.news/modeldrop/gemini-20-flash) (own drop) |
| 2 | Gemini 1.5 Pro | Google DeepMind | 85.4% |  | [Gemini 2.0 Flash](https://handyai.news/modeldrop/gemini-20-flash) |
| 3 | Gemini 1.5 Flash | Google DeepMind | 79.8% |  | [Gemini 2.0 Flash](https://handyai.news/modeldrop/gemini-20-flash) |

## OpenAI-MRCR

3 models from 1 drop.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | GPT-4.5 | OpenAI | 38.5 |  | [GPT-4.1](https://handyai.news/modeldrop/gpt-41) |
| 2 | GPT-4o | OpenAI | 31.9 |  | [GPT-4.1](https://handyai.news/modeldrop/gpt-41) |
| 3 | GPT-4.1 | OpenAI | 2 |  | [GPT-4.1](https://handyai.news/modeldrop/gpt-41) (own drop) |

## SWE-Bench (OpenHands)

3 models from 1 drop.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Nemotron 3 Nano | NVIDIA | 38.8% |  | [Nemotron 3 Nano](https://handyai.news/modeldrop/nemotron-3-nano) (own drop) |
| 2 | gpt-oss-20b | OpenAI | 34% |  | [Nemotron 3 Nano](https://handyai.news/modeldrop/nemotron-3-nano) |
| 3 | Qwen3-30B-A3B-Thinking-2507 | Alibaba | 22% |  | [Nemotron 3 Nano](https://handyai.news/modeldrop/nemotron-3-nano) |

## SWE-bench Pro v2 Hard

3 models from 1 drop.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Claude Opus 5.5 | Anthropic | 100% |  | [Mirendil, "Introducing Fern"](https://mirendil.com/news/introducing-fern/) |
| 2 | Fern 0.1 | Mirendil | 76.5% |  | [Mirendil, "Introducing Fern"](https://mirendil.com/news/introducing-fern/) (own drop) |
| 3 | Nemotron 3 Ultra | NVIDIA | 31.4% |  | [Mirendil, "Introducing Fern"](https://mirendil.com/news/introducing-fern/) |

## SWE-Lancer Diamond

3 models from 1 drop.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | GPT-4.5 | OpenAI | 32.6% |  | [GPT-4.5](https://handyai.news/modeldrop/gpt-45) (own drop) |
| 2 | GPT-4o | OpenAI | 23.3% |  | [GPT-4.5](https://handyai.news/modeldrop/gpt-45) |
| 3 | o3-mini | OpenAI | 10.8% | high | [GPT-4.5](https://handyai.news/modeldrop/gpt-45) |

## τ²-bench Airline

3 models from 1 drop.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | o3 | OpenAI | 64.8% |  | [GPT-5](https://handyai.news/modeldrop/gpt-5) |
| 2 | GPT-5 | OpenAI | 62.6% | high | [GPT-5](https://handyai.news/modeldrop/gpt-5) (own drop) |
| 3 | GPT-4.1 | OpenAI | 56% |  | [GPT-5](https://handyai.news/modeldrop/gpt-5) |

## AA-Omniscience

2 models from 1 drop.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Gemini 4 Carbon | Google DeepMind | 54 |  | [Gemini 4 Carbon](https://handyai.news/modeldrop/gemini-4-carbon) (own drop) |
| 2 | Claude Opus 5.5 | Anthropic | 48 |  | [Gemini 4 Carbon](https://handyai.news/modeldrop/gemini-4-carbon) |

## Artifacts Bench

2 models from 1 drop.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Claude Haiku 4.5 | Anthropic | 36.6% |  | [MAI-Code-1-Flash](https://handyai.news/modeldrop/mai-code-1-flash) |
| 2 | MAI-Code-1-Flash | Microsoft | 36.4% |  | [MAI-Code-1-Flash](https://handyai.news/modeldrop/mai-code-1-flash) (own drop) |

## BFCL V2 Live

2 models from 1 drop.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Llama 3.1 Nemotron Ultra | NVIDIA | 74.1% |  | [Llama 3.1 Nemotron Ultra](https://handyai.news/modeldrop/llama-31-nemotron-ultra) (own drop) |
| 2 | Llama 3.1 405B | Meta | 58.7% |  | [Llama 3.1 Nemotron Ultra](https://handyai.news/modeldrop/llama-31-nemotron-ultra) |

## Code refactoring tasks

2 models from 1 drop.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | GPT-5-Codex | OpenAI | 51.3% |  | [GPT-5-Codex](https://handyai.news/modeldrop/gpt-5-codex) (own drop) |
| 2 | GPT-5 | OpenAI | 33.9% | high | [GPT-5-Codex](https://handyai.news/modeldrop/gpt-5-codex) |

## Code review comments per PR

2 models from 1 drop.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | GPT-5 | OpenAI | 1.3 | high | [GPT-5-Codex](https://handyai.news/modeldrop/gpt-5-codex) |
| 2 | GPT-5-Codex | OpenAI | 0.9 |  | [GPT-5-Codex](https://handyai.news/modeldrop/gpt-5-codex) (own drop) |

## Code review high-impact comments

2 models from 1 drop.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | GPT-5-Codex | OpenAI | 52.4% |  | [GPT-5-Codex](https://handyai.news/modeldrop/gpt-5-codex) (own drop) |
| 2 | GPT-5 | OpenAI | 39.4% | high | [GPT-5-Codex](https://handyai.news/modeldrop/gpt-5-codex) |

## Code review incorrect comments

Lower is better. 2 models from 1 drop.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | GPT-5-Codex | OpenAI | 4.4% |  | [GPT-5-Codex](https://handyai.news/modeldrop/gpt-5-codex) (own drop) |
| 2 | GPT-5 | OpenAI | 13.7% | high | [GPT-5-Codex](https://handyai.news/modeldrop/gpt-5-codex) |

## FActScore

Lower is better. 2 models from 1 drop.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Grok 4.1 | SpaceXAI | 3% |  | [Grok 4.1](https://handyai.news/modeldrop/grok-41) (own drop) |
| 2 | Grok 4 Fast | SpaceXAI | 9.9% | non reasoning | [Grok 4.1](https://handyai.news/modeldrop/grok-41) |

## Hallucination Rate

Lower is better. 2 models from 1 drop.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Grok 4.1 | SpaceXAI | 4.2% |  | [Grok 4.1](https://handyai.news/modeldrop/grok-41) (own drop) |
| 2 | Grok 4 Fast | SpaceXAI | 12.1% | non reasoning | [Grok 4.1](https://handyai.news/modeldrop/grok-41) |

## Hermes Bench

2 models from 1 drop.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Nemotron 3.5 Super VL | NVIDIA | 63.4% |  | [Nemotron 3.5 Super VL](https://handyai.news/modeldrop/nemotron-35-super-vl) (own drop) |
| 2 | Nemotron 3 Ultra | NVIDIA | 56.3% |  | [Nemotron 3.5 Super VL](https://handyai.news/modeldrop/nemotron-35-super-vl) |

## Irregular medium cyber challenges

2 models from 1 drop.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | GPT-5 | OpenAI | 11 | xhigh | [GPT-5.1-Codex-Max](https://handyai.news/modeldrop/gpt-51-codex-max) |
| 2 | GPT-5.1-Codex-Max | OpenAI | 9 | xhigh | [GPT-5.1-Codex-Max](https://handyai.news/modeldrop/gpt-51-codex-max) (own drop) |

## MGSM

2 models from 1 drop.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | o1-preview | OpenAI | 90.8% |  | [o1](https://handyai.news/modeldrop/o1) |
| 2 | o1 | OpenAI | 89.3% |  | [o1](https://handyai.news/modeldrop/o1) (own drop) |

## OpenAI PRs

2 models from 1 drop.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | GPT-5.1-Codex-Max | OpenAI | 53% | xhigh | [GPT-5.1-Codex-Max](https://handyai.news/modeldrop/gpt-51-codex-max) (own drop) |
| 2 | GPT-5.1 | OpenAI | 45% | xhigh | [GPT-5.1-Codex-Max](https://handyai.news/modeldrop/gpt-51-codex-max) |

## RULER (128K)

2 models from 1 drop.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Nemotron Nano 2 | NVIDIA | 78.9% | reasoning off | [Nemotron Nano 2](https://handyai.news/modeldrop/nemotron-nano-2) (own drop) |
| 2 | Qwen3-8B | Alibaba | 74.1% |  | [Nemotron Nano 2](https://handyai.news/modeldrop/nemotron-nano-2) |

## RULER-100@1M

2 models from 1 drop.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Nemotron 3 Nano | NVIDIA | 86.3% |  | [Nemotron 3 Nano](https://handyai.news/modeldrop/nemotron-3-nano) (own drop) |
| 2 | Qwen3-30B-A3B-Thinking-2507 | Alibaba | 77.5% |  | [Nemotron 3 Nano](https://handyai.news/modeldrop/nemotron-3-nano) |

## Sycophancy rate

Lower is better. 2 models from 1 drop.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Grok 4 | SpaceXAI | 7% |  | [Grok 4.1](https://handyai.news/modeldrop/grok-41) |
| 2 | Grok 4.1 | SpaceXAI | 19% |  | [Grok 4.1](https://handyai.news/modeldrop/grok-41) (own drop) |

## UK AISI "The Last Ones" cyber range

2 models from 1 drop.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Claude Mythos Preview | Anthropic | 22 |  | [Claude Mythos Preview](https://handyai.news/modeldrop/claude-mythos-preview) (own drop) |
| 2 | Claude Opus 4.6 | Anthropic | 16 |  | [Claude Mythos Preview](https://handyai.news/modeldrop/claude-mythos-preview) |

## Video-MME

2 models from 1 drop.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | GPT-4.1 | OpenAI | 72% |  | [GPT-4.1](https://handyai.news/modeldrop/gpt-41) (own drop) |
| 2 | GPT-4o | OpenAI | 65.3% |  | [GPT-4.1](https://handyai.news/modeldrop/gpt-41) |

## WMDP Cyber

2 models from 1 drop.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Grok 4 | SpaceXAI | 79% |  | [Grok Code Fast 1](https://handyai.news/modeldrop/grok-code-fast-1) |
| 2 | Grok Code Fast 1 | SpaceXAI | 62.1% |  | [Grok Code Fast 1](https://handyai.news/modeldrop/grok-code-fast-1) (own drop) |

## CritPt v2 Hard

Unpublished research-level physics problems, hard subset, run by Artificial Analysis. 27 models from 0 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | GPT-6 Astra | OpenAI | 70.5% | max | [Artificial Analysis (2026-11-04)](https://artificialanalysis.ai/evaluations) |
| 2 | GPT-6.1 Sol | OpenAI | 69.5% | max | [Artificial Analysis (2026-11-04)](https://artificialanalysis.ai/evaluations) |
| 3 | Gemini 4 Carbon | Google DeepMind | 69% | high | [Artificial Analysis (2026-11-04)](https://artificialanalysis.ai/evaluations) |
| 4 | GPT-6.1 Terra | OpenAI | 68% | max | [Artificial Analysis (2026-11-04)](https://artificialanalysis.ai/evaluations) |
| 5 | Claude Opus 5.5 | Anthropic | 68% | max | [Artificial Analysis (2026-11-04)](https://artificialanalysis.ai/evaluations) |

22 more: https://handyai.news/api/benchmarks/critpt-v2-hard

## Terminal-Bench Science (standard harness)

Terminal-Bench Science run by a third party on one shared harness. 27 models from 0 drops.

| Rank | Model | Maker | Score | Setting | Source |
| --- | --- | --- | --- | --- | --- |
| 1 | Gemini 4 Carbon | Google DeepMind | 64.8% | high | [Artificial Analysis (2026-11-04)](https://artificialanalysis.ai/evaluations) |
| 2 | GPT-6 Astra | OpenAI | 63.3% | max | [Artificial Analysis (2026-11-04)](https://artificialanalysis.ai/evaluations) |
| 3 | Claude Opus 5.5 | Anthropic | 59% | max | [Artificial Analysis (2026-11-04)](https://artificialanalysis.ai/evaluations) |
| 4 | GPT-6.1 Sol | OpenAI | 58.1% | max | [Artificial Analysis (2026-11-04)](https://artificialanalysis.ai/evaluations) |
| 5 | GPT-6.1 Terra | OpenAI | 57.6% | max | [Artificial Analysis (2026-11-04)](https://artificialanalysis.ai/evaluations) |

22 more: https://handyai.news/api/benchmarks/terminal-bench-science-standard-harness

## Releases by month

- 2024-12: o1, Gemini 2.0 Flash, DeepSeek V3
- 2025-01: DeepSeek R1, o3-mini
- 2025-02: Grok 3, Claude 3.7 Sonnet, GPT-4.5
- 2025-03: Gemini 2.5 Pro
- 2025-04: Llama 4 Maverick & Scout, Llama 3.1 Nemotron Ultra, GPT-4.1, o3 & o4-mini, Gemini 2.5 Flash, Qwen3
- 2025-05: Mistral Medium 3, Claude Opus 4 & Sonnet 4, DeepSeek R1 0528
- 2025-06: o3-pro
- 2025-07: Grok 4, Kimi K2, GLM-4.5
- 2025-08: Claude Opus 4.1, gpt-oss-120b & gpt-oss-20b, GPT-5, Nemotron Nano 2, DeepSeek V3.1, Grok Code Fast 1
- 2025-09: GPT-5-Codex, Grok 4 Fast, Qwen3-Max, Claude Sonnet 4.5
- 2025-10: Claude Haiku 4.5, MiniMax-M2, Composer
- 2025-11: Kimi K2 Thinking, GPT-5.1, Grok 4.1, Gemini 3 Pro, GPT-5.1-Codex-Max, Claude Opus 4.5
- 2025-12: DeepSeek V3.2, GPT-5.2, Nemotron 3 Nano, Gemini 3 Flash, GPT-5.2-Codex
- 2026-01: Kimi K2.5
- 2026-02: Claude Opus 4.6, GPT-5.3-Codex, GLM-5, Qwen3.5, Claude Sonnet 4.6, Gemini 3.1 Pro
- 2026-03: GPT-5.4, Nemotron 3 Super, Composer 2, GLM-5.1
- 2026-04: Gemma 4, Claude Mythos Preview, Muse Spark, Claude Opus 4.7, GPT-Rosalind, GPT Image 2, Kimi K2.6, GPT-5.5, DeepSeek V4
- 2026-05: GPT-Realtime-2, Composer 2.5, Gemini 3.5 Flash, Gemini Omni, Claude Opus 4.8
- 2026-06: MAI-Code-1-Flash, Nemotron 3 Ultra, Fable 5 / Mythos 5, Kimi K2.7 Code, GLM 5.2, GPT-5.6 Sol, Terra, & Luna, Claude Sonnet 5
- 2026-07: Muse Spark 1.1, Grok 4.5, Inkling, Kimi K3, Qwen3.8-Max, Gemini 3.6 Flash, Claude Opus 5, MAI-Cyber-1-Flash
- 2026-08: Muse Spark 1.2, Nemotron 3.5 Lightning, Grok 4.6, MAI-Thinking-1, Qwen3.8, Gemini 3.7 Flash, GLM-5.3
- 2026-09: Fable 5.1, Gemini 3.8 Flash, Muse Spark 1.3, GPT-6 Astra, Grok 4.7, MiMo-V2.6-Pro, Claude Opus 5.5, GPT-6 Sol & GPT-6 Luna, Claude Sonnet 5.5, GPT-6.1 Sol
- 2026-10: Gemini 4 Argon, Mistral Large 4, Claude Haiku 5.5, Muse Wonder, Nemotron 3.5 Super VL, Muse Spark Open, DeepSeek V4.1 Flash 1017, MiniMax M3.1 Flash, Kimi K3.1 Pro, Nanbeige4.3-33B, GPT-6.1 Terra, Fern 0.1, Ling-3.1-flash-1019, LFM3-1.7B, Clover-35B-A3B-Preview, Particle-1, Gemini 4 Carbon, SenseNova-Loop-9B-Preview
- 2026-11: LongCat-2.5, Hy4