Hedge-Bench Research Coverage
Finance & Accounting · 2026-06-02
Share of Hedge-Bench's hidden analyst-derived rubric moves an agent's report covers with valid citations, credited by a majority of a three-model jury and averaged over three trials on 92 environments.
Top models (higher is better)
| Model | Score |
|---|---|
| GPT-5.6 Sol | 59.2 |
| Grok 4.6 | 58.3 |
| Claude Opus 5 | 55.9 |
| Claude Fable 5.1 | 55.9 |
| GPT-5.6 Terra | 50.9 |
| GLM-5.3 | 47.9 |
| Muse Spark 1.3 | 47.0 |
| Kimi K3 | 42.9 |
| DeepSeek V4 Pro 0813 | 37.3 |
| Gemini 3.8 Flash | 32.4 |