Anthropic’s latest Claude releases, including Opus 5.5 in September 2026 and Fable 5.1 earlier that month, have driven the highest reported HLE scores to 65% or above on multiple leaderboards, with tool-augmented runs reaching 67.7% for Opus 5.5. These gains stem from improved reasoning, adaptive tool use, and larger context windows on the 2,500-question benchmark spanning math, physics, and expert domains. Claude variants currently lead or tie most public evaluations ahead of OpenAI’s GPT-6 Astra and Google’s Gemini models, though scores vary by protocol (text-only versus tool-enabled) and subset. Further Claude updates or refined agentic setups before year-end remain the main catalysts that could shift the final 2026 peak.