OpenAI: Agents Solve the Navier–Stokes Millennium Problem

A swarm of agents running an unreleased next-generation model produced a proof of one of the deepest open questions in mathematics — published with a Lean formalization, and contested within hours over credit and data.

OpenAI said Monday that a team of its agents solved the Navier–Stokes existence and smoothness problem, one of the seven Millennium Prize Problems and a frontier of fluid dynamics that has resisted proof for decades. The company published a paper and a formal proof in the Lean proof assistant, and framed the result as evidence that a model "significantly more capable than GPT-6 Astra" can now pursue research goals with minimal human steering.

The celebration was short-lived. Two mathematicians — Levent Alpöge and Tristan Buckmaster — had spent nearly a year attacking the same problem with Claude and Codex. OpenAI congratulated them and denied accessing their work or any "specific user data," but acknowledged it cannot rule out that de-identified data was used to improve its models. The episode has turned a mathematical breakthrough into a live argument over who gets credit when frontier models and human researchers race toward the same answer.

Mistral Raises €3B, Europe's Largest Tech Equity Round

Three years after launch, a record round to scale training and inference compute.

Mistral closed a €3 billion Series D — the largest equity round ever raised by a European tech company — to accelerate its open-model research and its sovereign-AI ambitions.

The raise drew congratulations across the field and was read as a landmark bet on a homegrown European frontier lab.

AlphaGenome Atlas Maps All 9 Billion DNA Variants

An AI-powered, searchable database predicts the impact of every single-letter DNA change — free for academic researchers.

DeepMind's AlphaGenome Atlas lets researchers search the predicted effect of all ~9 billion possible single-letter DNA changes in an ordinary browser, no coding required. Google's Sundar Pichai called it an interactive resource and said he is "excited for the discoveries to come."

Astra Rolls Out to Codex and ChatGPT Work

Astra is now fully available to Plus, Pro, Business, and Enterprise users across Codex and ChatGPT Work, bringing the model directly into coding and workflow surfaces.

This week OpenAI has shipped Astra and Images 2.5 back-to-back, an unusually dense product cadence.

Terence Tao: Good Problems Are Being Mined Out

In a thread reacting to the Navier–Stokes announcement, Terence Tao argues that valuable open mathematical problems are a non-renewable resource. The point seems counterintuitive, since the number of possible questions is infinite. His analogy: a region can be short of water while surrounded by ocean. Mathematics can pose open problems at will — say, computing the 10^10^10th digit of pi — but most are not worth studying, unlikely to reveal deeper insight or connections, and either too easy for current tools or too hard to learn from.