Anthropic’s Claude Fable 5.2 recently topped independent and provider-reported runs of Humanity’s Last Exam at 74.3% on the AA-HLE subset and 74.2% overall, widening the gap over Moonshot’s Kimi K3.1 Pro at 70% with tools and earlier Claude Opus 5.5 releases near 61-67%. The 2,500-question benchmark, developed by the Center for AI Safety and Scale AI, tests expert-level reasoning across disciplines and continues to show steep gains as frontier large language models incorporate longer context, improved chain-of-thought, and agentic tool use. Competitive dynamics favor labs releasing frequent iterations—Anthropic and OpenAI have driven most leaderboard jumps this fall—while verification variants like HLE-Diamond and HLE-Verified introduce scoring differences that traders watch for resolution clarity. With December 31, 2026 as the cutoff, additional model drops or benchmark refinements remain key swing factors before year-end.
This market will resolve to "Yes" if any model achieves at least the specified accuracy on Humanity’s Last Exam by December 31, 2026, 11:59 PM ET. Otherwise, this market will resolve to "No". For resolution, “accuracy” refers solely to the value labeled “HLE Accuracy” or a clear equivalent metric if the data’s presentation or terminology is restructured, regardless of the model’s Calibration Error or any other displayed metric. The resolution source will be the official Humanity’s Last Exam leaderboard at https://agi.safe.ai/. If the resolution source becomes unavailable during the listed timeframe, this market will remain open to allow the relevant data to become available again. If the source remains unavailable after the end of the listed timeframe or is otherwise confirmed to be permanently unavailable, official Humanity’s Last Exam results published elsewhere may be used. If no official alternative source is available, this market will resolve to "No".