Recent releases have driven trader focus toward Anthropic and OpenAI models on Humanity’s Last Exam, a 2,500-question expert benchmark testing frontier reasoning resistant to simple search. As of early October 2026, Anthropic’s Claude Fable 5.1 and Mythos 5.1 lead standard closed-book evaluations at 65%, with Claude Opus 5.5 at 61–67.7% when tools are allowed. OpenAI’s GPT-6 Astra set a notable mark in September by scoring 82.9% on the refined HLE-Diamond subset with web and code tools, shifting market-implied odds higher for year-end thresholds above 75%. Competitive pressure from Google’s Gemini variants and open-weight releases like DeepSeek models adds uncertainty, while typical release cycles suggest further gains possible before December 31.