AIMultiple - FinanceReasoning (Hard): leaderboard
AIMultiple's run of the FinanceReasoning benchmark (Tang et al.) hard subset: 238 multi-step quantitative finance questions with free-form chain-of-thought answers, extracted by Claude Sonnet 4.5 and scored against ground truth with a 0.2% relative tolerance. Models run through provider APIs; accuracy %, higher is better.
Metric: Accuracy (%, 238 hard FinanceReasoning questions). Source: aimultiple.com. Saturation forecast: Estimated already saturated. 96 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 5.5 | 91.6 | #3 |
| 2 | Muse Wonder | 91.6 | #18 |
| 3 | Gemini 4 Argon | 90.76 | #9 |
| 4 | GPT-5.6 Pro Sol | 90.76 | #21 |
| 5 | GPT-5.6 Sol | 90.34 | #22 |
| 6 | Claude Fable 5 | 90.34 | #20 |
| 7 | Grok 4.6 | 90.34 | #29 |
| 8 | Claude Fable 5.1 | 90.34 | #10 |
| 9 | GPT-6.1 Pro Sol | 90.34 | #11 |
| 10 | Claude Opus 5 | 89.92 | #16 |
| 11 | GPT-6.1 Sol | 89.92 | #7 |
| 12 | Hy4 preview | 89.92 | #45 |
| 13 | GLM-5.3 | 89.5 | #44 |
| 14 | Kimi K3.1 Pro | 89.5 | #17 |
| 15 | Claude Opus 4.8 | 89.08 | #39 |
Interactive version: theaggregate.ai/benchmark?slug=aimultiple-financereasoning-hard · How It Works · Data refreshed daily, snapshot 2026-11-05.