Recent model releases have driven scores on Humanity’s Last Exam (HLE) into the low-to-mid 60% range on standard leaderboards, with Anthropic’s Claude Opus 5.5 and Fable 5.1 posting the highest verified results near 61–65% depending on the evaluation harness. Tool-augmented runs and the refined HLE-Diamond subset have produced even higher marks, including OpenAI’s GPT-6 Astra reaching 82.9% with web and code tools, though these use narrower or modified question sets. Competitive pressure from OpenAI, Google, and Meta continues, yet the benchmark’s expert-curated questions across math, science, and humanities remain far from saturated. Traders are watching for any additional frontier releases or official leaderboard updates through year-end that could push the top score across key thresholds before December 31.