An unreleased research version of Claude improved a decades old mathematical bound related to the Riemann hypothesis, Anthropic said in a research post Monday.
Working through Claude Code across two sessions, the model coordinated about 60 Claude subagents that ran thousands of numerical checks, raising a lower bound for the fraction of zeros of the Riemann zeta function known to satisfy the Riemann hypothesis from 41.6% to 67.2%, according to Anthropic.
The advance came from treating positive and negative definiteness together across the full space and allowing the quadratic form used in the proof to be non diagonal, Anthropic said, rather than from new mathematics developed from scratch.
Anthropic said it does not expect the techniques Claude used to lead to a proof of the Riemann hypothesis itself, describing the result as progress on a related, narrower problem. The company said the result was checked by outside mathematicians and includes a formally verifiable proof.
For builders watching what agentic systems can do in domains with a checkable ground truth, this is a data point worth reading with its own caveats attached: coordinated subagents plus large scale numerical search moved a real, decades static bound, but Anthropic's own framing keeps that well short of the unsolved conjecture that makes headlines.