**Anthropic’s Claude Fable 5.1 and Opus 5 currently lead HLE leaderboards at 59–65% (text-only or tool-augmented variants as of mid-September 2026), while OpenAI’s strongest reported results sit at 54.7–58.7% (GPT-6 Astra, GPT-5.4 Pro, GPT-5.5 Pro).** This gap reflects Anthropic’s edge in adaptive reasoning, tool integration, and long-horizon task performance on the 2,500-question benchmark covering expert-level math, physics, biology, and other domains. Scores have risen rapidly in 2026 through better chain-of-thought, thinking modes, and agentic scaffolding, moving the frontier from low teens percent early in the year into the mid-50s. OpenAI’s releases have kept pace but trail slightly on the hardest closed-ended academic questions, where verification and calibration remain challenging. Trader consensus prices 50%+ and 55%+ thresholds very high, reflecting expected continued scaling and model iterations, while 65%+ and above carry lower odds due to the competitive lead and typical product timelines. Key upcoming catalysts include OpenAI’s next flagship releases or reasoning updates before year-end, potential tool-use or agent improvements, and any new independent evaluations. HLE remains unsaturated, so incremental gains from training advances or post-training techniques could still shift relative positioning quickly.