Anthropic's frontier Claude models, including Opus 5.5 and Fable 5.1, currently lead Humanity's Last Exam leaderboards with scores of 59-67% as of mid-October 2026, depending on whether tool use or verified subsets apply. This marks substantial gains from single-digit performance in early 2025, driven by iterative improvements in large language model reasoning, chain-of-thought techniques, and multi-domain expert knowledge across the benchmark's 2,500 closed-ended questions. OpenAI's GPT-6 variants and Google's Gemini releases trail closely but show narrower gains, while open-weight models like DeepSeek lag further behind. Ongoing model releases and potential benchmark refinements, such as HLE-Diamond subsets, remain key variables that could shift the highest score before year-end.