Open FinLLM Reasoning: leaderboard
Metric: Accuracy (%). Source: huggingface.co. 26 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | DeepSeek V3 | 61.3 |
| 2 | GPT-4o | 61.01 |
| 3 | DeepSeek R1 | 60.87 |
| 4 | GPT-4.5 | 60.43 |
| 5 | DeepSeek R1 Distill Llama 70B | 59.27 |
| 6 | DeepSeek R1 Distill Qwen 32B | 57.4 |
| 7 | Qwen 2.5 32B Instruct | 56.17 |
| 8 | Llama 3.3 70B Instruct | 56.04 |
| 9 | O1 | 54.05 |
| 10 | Qwen 2.5 72B Instruct | 53.71 |
| 11 | DeepSeek R1 Distill Qwen 14B | 53.18 |
| 12 | QwQ-32B | 52.91 |
| 13 | Qwen 2.5 14B Instruct | 52.72 |
| 14 | Llama 3.1 70B Instruct | 52.21 |
| 15 | Qwen2.5-Math-72B-Instruct | 50.02 |
Interactive version: theaggregate.ai/benchmark?slug=open-finllm-reasoning · How It Works · Data refreshed daily, snapshot 2026-09-05.