BizBench - CodeFinQA: leaderboard
Metric: Accuracy (%; 844 FinQA questions over report text and tables answered by generating Python code, answer within 1% of the reference, 3-shot). Source: arxiv.org. Saturation forecast: Estimated already saturated. 16 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-4 | 78.8 |
| 2 | GPT-3.5 | 67.5 |
| 3 | Mixtral 8x7B | 58.5 |
| 4 | Llama 2 70B | 57.3 |
| 5 | Mistral 7B | 48.8 |
| 6 | Llama 2 13B | 33.4 |
| 7 | starcoder | 31.2 |
| 8 | mpt-30B | 31 |
| 9 | Llama 2 7B | 21.9 |
| 10 | falcon-40B | 18.4 |
| 11 | mpt-7B | 6.6 |
| 12 | falcon-7B | 2 |
Interactive version: theaggregate.ai/benchmark?slug=bizbench-codefinqa · How It Works · Data refreshed daily, snapshot 2026-09-26.