Recent surges in AI mathematical reasoning, driven by models like OpenAI’s GPT-6 Astra and Google DeepMind’s Gemini multi-agent systems (including AlphaProof Nexus), have accelerated solutions to research-level problems. Epoch AI’s FrontierMath: Open Problems benchmark—focusing on verifiable, previously unsolved conjectures—has seen multiple AI-assisted resolutions added in 2026, including Erdős problems and number theory results, with formal Lean verifications and arXiv preprints crediting frontier LLMs. Competitive releases from OpenAI, Anthropic, and Google, plus multi-agent frameworks handling proof search and verification, have shifted trader sentiment toward continued progress. Key near-term catalysts include Epoch AI’s ongoing Open Problems updates and new model benchmarks through late 2026, though full community peer review and official classifications remain resolution thresholds.