Hedge-Bench - Risk: leaderboard
Metric: Macro dense mean rubric score (out of 4) on the Risk task category: each trial scores 0 to 4 by the themes it covers, 4 only with full coverage plus a synthesis that reconciles conflicting evidence, averaged within each task and then over the tasks where the model had a valid run; 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 June 2028. 8 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Claude Sonnet 4.6 | 1.75 |
| 2 | Claude Opus 4.7 | 1.58 |
| 3 | Claude Opus 4.8 | 1.51 |
| 4 | Gemini 3.5 Flash | 1.31 |
| 5 | GPT-5.5 | 1.3 |
| 6 | Gemini 3.1 Pro (Preview) | 0.95 |
| 7 | Claude Haiku 4.5 | 0.92 |
| 8 | GPT-5.4 Mini | 0.53 |
Interactive version: theaggregate.ai/benchmark?slug=hedge-bench-risk · How It Works · Data refreshed daily, snapshot 2026-09-29.