OpenAI这是拿千禧年难题当Benchmark刷啊。。。

TL;DR - OpenAI says a system of thousands of AI agents made substantial progress on Millennium Prize problems, reportedly completing a Navier–Stokes proof and potentially targeting the Hodge conjecture next. The claims remain under review and raise unresolved questions about verification, research priority, data use, and credit for human mathematicians.

• OpenAI reportedly used about 10,000 parallel agents on Navier–Stokes, generating 2.7 million messages and 130 billion output tokens before GPT-6 Astra performed Lean formalization and verification.
• The broader experiment reportedly produced 4.9 million agent messages and roughly 300 billion output tokens across several unsolved problems.
• OpenAI states that recent Codex prompts from mathematician Tristan Buckmaster could not have influenced the internal model, but no public technical evidence independently verifies that claim.
• Reports that OpenAI has made progress on the Hodge conjecture—and that the internal model is code-named “Aeon”—remain unconfirmed rumors.