Claude Mythos reportedly solves OpenAI's landmark Erdős problem with a "cute, simple proof"

The more telling detail isn't the proof itself but the word 'independently': two labs converging on the same hard mathematical result in near-simultaneous fashion suggests the capability threshold for this class of problem was crossed at roughly the same time across the frontier, which implies the gap between Anthropic and OpenAI on formal reasoning may be narrower than either lab's marketing suggests.

This sits in a different layer than our recent coverage of OpenRouter's $1.3B valuation, which tracked infrastructure demand rather than raw model capability. That said, the OpenRouter story made a structural point worth connecting here: as enterprises treat model selection as a commodity decision, parallel breakthroughs like this one accelerate that dynamic. If Claude Mythos and OpenAI's model reach the same mathematical result independently, the differentiator for buyers shifts further toward cost, latency, and orchestration rather than which lab 'won' a given benchmark. The race to solve open problems becomes a proof-of-capability signal, not a durable moat.

Watch whether either lab submits the proof to peer-reviewed verification within the next 90 days. Independent mathematical confirmation (or a flaw found by the community) will determine whether this is a genuine research contribution or a well-structured pattern match on existing literature.