Frontier AI labs continue pushing mathematical reasoning capabilities, with MathArena evaluations showing leading models like Claude Opus 5 variants and GPT-5 series already posting scores above 80% on recent USAMO and AIME-style problems as of mid-2026. Anthropic and OpenAI releases have driven the strongest gains through scaled test-time compute and specialized training, narrowing gaps to near-saturation on established competition benchmarks while open-weight contenders like Kimi and GLM trail further behind. Trader sentiment reflects this momentum, tempered by the shift toward harder, uncontaminated problems where progress remains uneven. Key catalysts ahead include expected fall model updates, potential new IMO or AIME problem sets, and any disclosed internal benchmarks that could accelerate or stall consensus on whether thresholds are cleared before year-end.