Anthropic’s Claude Opus 5.5 and Fable 5.1 currently lead most verified Humanity’s Last Exam leaderboards at 61–65% as of early October 2026, reflecting gains from scaled reasoning training, tool integration, and targeted post-training on expert-level questions. OpenAI’s GPT-6 Astra has posted higher marks (up to 82.9%) on refined HLE-Diamond subsets with tools, while Google, Meta, and open-weight labs trail in the mid-50s to low-60s. Frontier scores have risen roughly 30 points in the past year, driven by rapid iteration cycles and benchmark refinements that reduce noise. With three months remaining, additional releases from Anthropic and OpenAI, plus further tool-use optimizations, represent the main catalysts that could push the year-end ceiling higher or expose saturation limits on the 2,500-question set.