Machine Learning

A research system built by Physical Superintelligence, a laboratory known as PSI, has produced a proof of the Aaronson-Ambainis conjecture, a central open problem in quantum computing that had stood since 2008. PSI announced the result on Friday, saying that its internally developed system, running instances of a non-public Anthropic model on PSI's own infrastructure, had a complete proof during the third day of the run. The proof was then accepted by Lean, a proof assistant that checks every logical step by computer.

Anthropic is now improving its AI models at about twice the pace it managed before late 2025, and most of that acceleration appears to come from the company using AI to help build AI. That is the central finding of an independent assessment published on November 5 by METR, a nonprofit that evaluates frontier AI systems, after its staff spent eleven weeks examining Anthropic's internal research process.

For a growing share of the world's software, the first draft no longer comes from a person typing at a keyboard. GitHub's annual Octoverse report, published on October 28, says that by September 2026 developers started more than 40 percent of pull requests with a coding agent's first draft, up from 17 percent in March. That is 110 million agent-started pull requests in September alone, about six times March's volume.

Anthropic says its Claude Mythos 5 model has designed quantum circuits that improve on the best previously published versions, in some cases by an order of magnitude or more. In a research post published on October 16, 2026, the company reported that Claude, working in a closed loop with an automated evaluator and largely on its own, took on eight computational primitives — the basic routines from which larger quantum algorithms are built — spanning chemistry, optimization, quantum simulation, number theory and numerical computing.

Google's race to catch up in AI-assisted coding may have a new contender — and it is not the model everyone is waiting for. Business Insider reported this week that Google employees are internally testing a Gemini 4 variant codenamed Carbon, and early feedback suggests it could be the strongest coding model the company has built.

CoreWeave built its reputation selling raw AI compute. Now it wants to sell the software layer on top of it. At its Fully Connected conference, the company unveiled CoreWeave Forge, a development platform that folds model serving, observability, post-training, and evaluation into a single environment — and it comes with a striking performance claim: a preview feature called RL Rollouts that improved model reload latency by 15x in testing.

French AI lab Mistral has launched Mistral Large 4, a 1.05-trillion-parameter natively multimodal model the company has nicknamed "le Chonk," now available in public preview via API. Open weights are promised by the end of October, after a red-teaming phase.

Anthropic released Claude Haiku 5.5 on Wednesday, the third launch in the Claude 5.5 family after Opus 5.5 and Sonnet 5.5 earlier this fall. Where the bigger siblings brought capability, Haiku 5.5 brings aggression: Anthropic calls it the cheapest, fastest, and most capable small model it has ever shipped, and the pricing backs up the claim.