OpenAI’s early October release of 722 papers detailing an internal frontier model’s solutions or advances on 377 open problems—including progress toward multiple Millennium Prize questions and variants of the Riemann hypothesis—has sharply raised expectations for further 2026 breakthroughs. This followed the company’s September claim of resolving a Navier-Stokes variant via thousands of parallel AI agents and Lean-verified proofs. Competing labs show similar momentum: DeepMind’s AlphaProof systems have cracked additional Erdős problems, while startups like doubleAI achieved low-cost gains on long-standing graph theory bounds. High FrontierMath Tier 4 scores above 90% for leading models such as GPT-6 variants underscore rapid capability gains, though mathematicians continue scrutinizing novelty, verification, and compute efficiency amid ongoing workshops and advisory reviews.