BizFinBench: leaderboard
BizFinBench v1 Chinese financial evaluation: the reported unweighted mean over nine business-oriented task scores, with GPT-4o-based IteraJudge where specified by the original experiment. These 25-model results match the v1 paper and are separate from the newer bilingual BizFinBench v2 dataset.
Metric: V1 reported nine-task mean score (0–100). Source: github.com. Saturation forecast: Estimated already saturated. 25 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | DeepSeek R1 | 73.05 | #251 |
| 2 | GPT-4o | 71.8 | #343 |
| 3 | DeepSeek V3 | 71.57 | #321 |
| 4 | Gemini 2.0 Flash | 69.75 | #342 |
| 5 | Qwen 3 32B | 68.26 | #443 |
| 6 | Qwen 2.5 72B Instruct | 67.7 | #454 |
| 7 | Qwen 3 14B | 67.05 | #537 |
| 8 | DeepSeek R1 Distill Qwen 32B | 66.29 | #655 |
| 9 | QwQ-32B | 65.77 | #421 |
| 10 | Claude 3.5 Sonnet | 65.59 | #350 |
| 11 | Llama 4 Scout | 61.17 | #656 |
| 12 | Qwen 3 4B | 59.94 | #829 |
| 13 | DeepSeek R1 Distill Qwen 14B | 59.49 | #844 |
| 14 | Qwen 2.5 VL 7B | 56.68 | #643 |
| 15 | Qwen 2.5 7B Instruct | 56.35 | #862 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=bizfinbench · How It Works · Data refreshed daily, snapshot 2026-10-20.