Anthropic's Claude rewrites math bound, boosts Riemann zero proof to 67.2%
Claude didn't crack Riemann, but raised the critical line bound to 67.2% using 60 sub-agents.
Anthropic's unreleased research version of Claude took on one of mathematics' greatest challenges, the Riemann Hypothesis. Though it failed to deliver a full proof, the model unexpectedly raised the established lower bound for the proportion of zeros on the critical line from a decades-long 41.6% to 67.2%. This is not a proof of the hypothesis itself, but a significant numerical improvement in a key related estimate, sparking intense discussion across the academic community.
The real breakthrough lies in how the AI worked. Claude did not rely on known answers but deployed approximately 60 sub-agents to form a kind of "virtual research team." This setup endured hundreds of failures, independently reviewed vast literature, and recombined classical mathematical techniques. Industry experts see this as a preview of future research workflows, where AI handles intensive trial-and-error and cross-referencing, drastically lowering the marginal search cost for humans exploring unknown territories.
- Claude increased the lower bound of zeros on the critical line from 41.6% to 67.2%, a milestone held for decades.
- The AI used a multi-agent framework with ~60 sub-agents acting as a virtual research team, enduring hundreds of failures.
- No Riemann Hypothesis proof was achieved, but the method demonstrates autonomous error verification and long-range collaboration.
Why It Matters
AI multi-agent systems can radically cut exploration costs, accelerating fundamental scientific breakthroughs.