Open FinLLM Reasoning - FinQA: leaderboard
Open FinLLM Reasoning run of FinQA, testing numerical reasoning over financial reports and tables.
Metric: Accuracy (%). Source: huggingface.co. Saturation forecast: Estimated already saturated. 26 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 2.5 72B Instruct | 73.38 | #454 |
| 2 | DeepSeek V3 | 73.2 | #321 |
| 3 | Qwen 2.5 32B Instruct | 73.11 | #498 |
| 4 | GPT-4o | 72.49 | #343 |
| 5 | Qwen2.5-Math-72B-Instruct | 69.74 | #717 |
| 6 | GPT-4.5 | 68.94 | #274 |
| 7 | Llama 3.3 70B Instruct | 68.15 | #534 |
| 8 | Qwen 2.5 14B Instruct | 67.44 | #647 |
| 9 | DeepSeek R1 Distill Llama 70B | 66.73 | #594 |
| 10 | DeepSeek R1 Distill Qwen 32B | 65.48 | #655 |
| 11 | DeepSeek R1 | 65.13 | #251 |
| 12 | DeepSeek R1 Distill Qwen 14B | 63.27 | #844 |
| 13 | Llama 3.1 70B Instruct | 63.18 | #561 |
| 14 | QwQ-32B | 61.22 | #421 |
| 15 | Llama 3 70B Instruct | 58.92 | #634 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=open-finllm-reasoning-finqa · How It Works · Data refreshed daily, snapshot 2026-10-20.