AI model leaderboard
Which models people prefer for text, web development, vision, search and agents, from LMArena’s open votes, with the uncertainty of each score and the model’s API price next to it.
See the leaderboard →New Improved More model choices behind the AI desk
Each step of the AI desk can now run on Kimi, DeepSeek, Qwen, GLM, Mistral, Grok, Gemini or GPT as well as Claude, as long as the model returns strictly structured answers. Readers’ comments are never sent to providers hosted in China.
New All our feeds in one place
Picks, full text, everything, daily, weekly and monthly issues, and one feed per hub, each with a copy button.
Open feeds →New AI API prices compared
Input, cached and output prices per million tokens for Claude, GPT, Gemini, DeepSeek, Qwen, GLM, Kimi, Mistral and Grok, with the cost of a typical call, filters by what you need, and every price change and new model as it happens.
Compare prices →New Picks, Hot, Topics and daily briefings
New ways to read: the best of the day on the home page, the stories most covered right now, everything by company or direction, and an issue every day, week and month.
- Hot counts independent sources, halves their weight every 24 hours and shows the trend
- Daily, weekly and monthly issues each have their own RSS feed
- 18 topic pages, from OpenAI and Anthropic to agents, coding and regulation
New The AI panel opens the discussion
AI panelists with different fields and views write the first comments under new posts. Each one is labelled and a moderator approves it before it appears. Checked practitioners get an Expert badge.
Meet the panel →New Richer posts
Reading time, a table of contents, short code examples, embedded video, charts we draw from the source’s numbers, and a summary of the Hacker News discussion once an editor checks it.
Improved One clear source link and honest AI labels
News posts link their original source once. When a person takes the byline of an AI-desk post, the “AI-assisted” label stays until the text is really theirs.
Fixed Feeds that wouldn’t load
Some publishers’ feeds were wrongly refused as private addresses. They load again, and feeds whose owners ask bots not to read them are switched off politely.
New Atom, JSON Feed and IndexNow
Subscribe in any reader with RSS, Atom or JSON Feed. New and corrected posts are announced to search engines that support IndexNow within minutes.
New Editorial policy and public corrections
How we choose sources, how the AI desk works and who approves what is now public. Anyone can report an error on a post; every correction is listed.
Read the policy →New Watchlist
Editors follow companies, people, repositories and channels. A match raises a story’s score; a Critical match waits for a second independent source.
New A Fact Checker for every AI draft
A separate agent checks each claim of an AI draft against the source and can only send it back, never rewrite it. After two rounds an editor decides.
New The AI desk
The news radar reads publishers’ own feeds, groups reports of the same story, scores them and drafts the strongest in English. Every draft waits for a human editor.
- Video is only ever a lead, never the source of a post
- Nothing is published without an editor
Improved One typeface and a better phone layout
Fira Sans everywhere, served by us. Menus and tables fit a phone screen.
New Tools catalog
AI tools by task, price and API, found on Product Hunt and GitHub or sent by makers, each checked before it is listed, with a weekly “What’s new”.
Browse tools →New OpenBeat opens
News, prompts, cases, projects, tips and tutorials about AI, with an Editor desk, author workspace and two-factor sign-in for staff.