Hedge-Bench - Valid Move Coverage: leaderboard

Metric: Valid moves covered (%), macro-averaged over tasks: mean within-theme share of expert moves hit, after discounting moves supported by hallucinated facts; 102 open-ended financial-analyst research tasks derived from recorded expert analyst discussions, run in the Harbor harness with the Terminus 2 agent over a sandboxed document corpus, eight trials per task, graded by a Gemini-3.1-Pro judge against expert rubric moves with hallucination-supported moves discounted; higher is better. Source: arxiv.org. Saturation forecast: Around November 2027. 8 models tracked.

Top models

#ModelScore
1Claude Sonnet 4.654.9
2Claude Opus 4.753.9
3Gemini 3.5 Flash49.8
4GPT-5.549.7
5Claude Opus 4.849.1
6Gemini 3.1 Pro (Preview)37.6
7Claude Haiku 4.536.9
8GPT-5.4 Mini26.1

Interactive version: theaggregate.ai/benchmark?slug=hedge-bench-valid-move-coverage · How It Works · Data refreshed daily, snapshot 2026-09-29.