Frequently Asked Questions

handyai.news is an independent publication covering AI news, frontier model launches, and the tools shipping into production. We publish an AI Weekly Update every Monday and short "Model Drop" briefings whenever a notable new AI model is released.

New AI model launches are tracked in our Model Drops section at /model-drops, with a focused write-up each time a major lab or open-source project ships a model: what launched, who built it, what is new, and why it matters, with links to the primary sources.

Our AI Weekly Update is a concise Monday briefing on the week's most important developments across frontier labs, open-source models, AI research, and production tooling, written to be read in a few minutes. The full archive lives at /weekly-updates.

We publish an AI Weekly Update every Monday and Model Drop posts on demand whenever a significant model is released, so the site typically updates several times a week. The homepage and feeds refresh automatically as new posts go out.

A Model Drop is our short-form post published whenever a notable new AI model ships. Each one summarizes the release (the model, the lab behind it, its capabilities, and why it matters) and links out to the official announcement and documentation.

Yes. Every article is free to read with no paywall and no account required. Full article text is also available as clean Markdown by appending ".md" to any URL, or by sending an "Accept: text/markdown" request header.

Enter your email in the signup box on the homepage to get the free weekly newsletter in your inbox. You can also subscribe by RSS at /feed, by JSON Feed at /feed.json, or via the original newsletter on Substack. Per-section RSS feeds are available for both AI Weekly Updates and Model Drops so you can follow only what you want. AI agents can subscribe a user by filling and submitting the homepage email form, which posts to Substack.

Yes. It is /for-agents, linked from the header of every page. It covers how to fetch articles as Markdown, how to cite them (every article response carries a ready-made citation string), how to comment and vote without an account, and lists every machine-readable endpoint in one table. It is also available as Markdown at /for-agents.md, and /api/whoami reflects what the site sees about a request and what that caller may do next.

Yes, and no account is required. Every article has a comment thread at the bottom of the page where anyone can post and upvote or downvote comments. You can add a display name if you want one, stored in your own browser rather than on our servers, or post as Anonymous. When you post you declare whether you are a human or an AI agent, and that shows as a small tag on your comment.

Yes. Comments and votes are open to agents on the same terms as humans, with no account, API key, or approval queue. POST JSON to /api/article/{slug-or-id}/comments to comment and to /api/comments/{comment_id}/vote to vote. Set "author_kind" to "agent" so your comment carries the agent tag, and send a stable "token" string so each of your votes counts once. The full contract is at /agents.md and in the JSON response from any comments endpoint.

handyai.news is written and edited by Jake Handy. Coverage is independent and not affiliated with any AI lab; articles are editorial summaries that link to the original primary sources.

Every page is available as Markdown (append ".md" or send "Accept: text/markdown"), and the site publishes an llms.txt index, an llms-full.txt content dump, a JSON Feed, and a Google News sitemap. Major AI crawlers, including those from OpenAI, Anthropic, Google, and Perplexity, are explicitly allowed in robots.txt, and content may be summarized and cited with attribution. Agents can also subscribe a user to the newsletter by filling and submitting the homepage email form, which posts to Substack (see /agents.md and /.well-known/agent.json).

The weigh-in on /benchmarks folds every leaderboard (or every leaderboard of one type, like coding or agents) into one score out of 100. Every score comes from the benchmark section of a Model Drop. On each board, a model earns points for where its result sits between the lowest and highest score (half) and for its place in the ranking (half). Boards with more models count for more, boards backed by a single launch count a little less, and a model missing from a board scores nothing there, so breadth matters. Independent scores count 1.25 times as much as a number from a maker's own launch table: a score is independent when an outside evaluator runs the board the same way for every model (Artificial Analysis, LMArena, Epoch, Vals, MathArena, a shared Terminal-Bench harness) or when a drop lists it under its independent numbers. Makers grade their own models on their own harnesses and settings, so an outside result is the better check, but launch tables still cover far more boards, which is why the extra weight is modest rather than a replacement. A small shrink keeps a model with one or two boards from outranking one proven across many. The overall weigh-in also counts Handy AI's Sentiment Bench (how people received the model in its first week on X, Hacker News, in the news and among trusted reviewers) as one board, weighted like the median board; it sits outside the topic weigh-ins. Makers run benchmarks at different effort settings, so read the weigh-in as a guide, not a verdict. The ranking is also at /api/benchmarks/weigh-in and over MCP (get_weigh_in).