Anthropic released Claude Haiku 5.5 on Wednesday, the third Claude 5.5 model in about a month, priced at $0.10 per million input tokens and $0.50…
OpenAI released 372 math results from a frontier model it has not made public, written up as 722 manuscripts on GitHub; The New York Times…
OpenAI expanded the GPT-6 lineup with Sol and Luna at half prior GPT-5.6 promotional API rates, saying it trained both with methods similar to flagship…
First Claude 5.5 family model claims Fable 5.1-level performance on most work at about 40% less cost than Opus 5, with $4/$20 per million token…
Company post and The Verge: up to 50% lower latency; Flare and Sunburst hit the API; Sketch turns doodles into prompts.
TechCrunch, Axios, and Engadget: first licensed-label lineup; older scrape-trained models retire.
Company post, NYT, and New Scientist: unnamed model beyond Astra; Lean proof; Clay prize not claimed; concurrent Buckmaster–Alpöge work disputed.
Nature and DeepMind: precomputed effects for every single-letter human mutation; AVI impact score; free non-commercial portal.
Company research post: ~13M lines of Lean over 11 days; Kevin Buzzard calls it an extraordinary autoformalization achievement.
The Verge and TechCrunch: hourly 5-kilometer AI forecasts from live satellite data. Google claims sharper rain calls. WindBorne disputes the “first” line.