Anthropic’s Claude Opus 5.5 and Fable 5.1 variants currently post the strongest results on standard Humanity’s Last Exam evaluations, reaching 61.4% on Artificial Analysis’s text-only subset and up to 65% on select leaderboards as of mid-October 2026. These scores reflect iterative gains in expert-level reasoning across 2,500 multi-modal academic questions, outpacing OpenAI’s GPT-6 Astra (around 55% no-tools) and Google’s Gemini models on most independent runs. Tool-augmented protocols produce markedly higher figures, including 82.9% for GPT-6 Astra on the September HLE-Diamond subset, highlighting sensitivity to evaluation settings. Frontier labs continue releasing updated models with improved chain-of-thought and agentic capabilities before year-end, sustaining the roughly 30-point annual climb observed since early 2025 baselines. Market participants weigh these protocol differences and upcoming releases when assessing the year’s peak reported score.