Anthropic’s Claude models currently lead Humanity’s Last Exam (HLE) leaderboards, with Claude Opus 5.5 and Fable 5.1 variants posting the highest verified scores of 61–65% on the core 2,500-question benchmark and up to 67.7% in tool-augmented settings as of early October 2026. This reflects sustained gains from single-digit results in early 2025, driven by iterative scaling, improved reasoning chains, and selective tool use such as web search and code execution. OpenAI’s GPT-6 Astra has demonstrated outlier performance on refined HLE-Diamond subsets (reaching 82.9% with tools), while Google’s Gemini and Meta’s Muse Spark series trail but remain competitive. Traders are watching for further frontier releases and any official leaderboard updates before year-end, as benchmark saturation remains low and small methodological differences can shift the highest recorded accuracy.