Six hours after OpenAI, Anthropic says Claude broke the same math hypothesis — on Oct 16, with 'no human input'

"An internal Anthropic model also refutes the ETH. With only 14m output tokens and already on October 16th. Suddenly everything at once," AI news account @kimmonismus posted on Oct 27 (581 likes). About six hours after OpenAI said it had disproved the Exponential Time Hypothesis, Anthropic put out a paper claiming the same thing.

The paper, "A 2^{o(n)} Time Algorithm for 3-SAT," lists a single author, "Claude (Anthropic)," on the Hexagon preprint site and in Anthropic's formal-math GitHub repo. "The core algorithm was found on October 16 by an internal Claude research model with 14M output tokens and no human input," its provenance statement reads. "Anthropic researchers worked with Claude to improve the presentation." It adds that an OpenAI internal model "independently and concurrently" claims a similar result.

The paper describes a deterministic algorithm that solves 3-SAT on n variables in subexponential time and even counts the satisfying assignments. The proof was formalized in Lean 4 by AI models and re-checked by two independently written kernels. Anthropic is careful to add that the result "is purely asymptotic and says nothing about solving formulas in practice." The repo's README also claims Claude Opus 5.5 at high effort can rederive the refutation from a few-line condensed description.

Anthropic mathematician Levent Alpöge (@__alpoge__), who surfaced the paper, drew 359,000 views and said he claims "absolutely zero credit." Posts claiming both labs' models hit the same bound on the same day (@deedydas) spread quickly. On the timing, @apples_jimmy quipped that "you don't have OpenAI employees crying on x about Anthropic 'stealing' their idea" (329 likes). The open examination organized by MIT's Ryan Williams and colleagues currently covers OpenAI's claim.