Claude formalizes Fermat’s Last Theorem in Lean
Anthropic says Claude agents produced the first end to end, computer checked FLT proof in Lean over 11 days — about 13M lines of code and 29,500 intermediate theorems. Human input was mostly high lev…
News
What changed in AI tooling this week: releases, pricing, outages, acquisitions. Every item links to its source.
Google open-sources Mantis for agentic bug finding and fixing
Google released Mantis, an open source agent harness that finds, sandboxes reproduces, and patches vulnerabilities — aimed at cutting hallucinated findings from naive AI code scanners.
Astra’s 99.9% ARC score came from the harness, not the model
ARC Prize ran GPT 6 Astra two ways: 99.9% under OpenAI’s Provider Adapter harness, 62.7% under the benchmark’s standard harness. Same weights, different scaffolding — and ARC Prize still says it isn’…
OpenAI ships GPT-6 Astra
OpenAI launched GPT 6 Astra for ChatGPT and API users, claiming frontier scores on computer use, coding, and math. Rollout starts limited, then expands to Plus/Pro/Business/Enterprise.