BizBench: leaderboard
BizBench: Evaluates financial analysis, accounting, market reasoning, and quantitative business tasks.
Metric: Mean accuracy (%) over the eight BizBench tasks (FinKnow, FormulaEval, ConvFinQA and TAT-QA extraction, SEC-Num, FinCode, CodeFinQA, CodeTAT-QA), each in its fixed zero- or few-shot setting. Source: huggingface.co. Source: arxiv.org. Saturation forecast: Estimated already saturated. 15 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-4 | 84.6 | #438 |
| 2 | GPT-3.5 | 72.5 | #818 |
| 3 | Mixtral 8x7B | 69.9 | #1143 |
| 4 | Llama 2 70B | 68.4 | #1201 |
| 5 | Mistral 7B | 61.8 | #1397 |
| 6 | Llama 2 13B | 54.7 | #1436 |
| 7 | mpt-30B | 53.5 | #1476 |
| 8 | Llama 2 7B | 48.2 | #1520 |
| 9 | starcoder | 47.5 | #1577 |
| 10 | falcon-40B | 41.3 | #1442 |
| 11 | mpt-7B | 37.1 | #1619 |
| 12 | falcon-7B | 28 | #1598 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=bizbench · How It Works · Data refreshed daily, snapshot 2026-10-20.