BizBench - FinKnow: leaderboard
FinKnow, a BizBench task (Koncel-Kedziorski et al., 2023): 877 multiple-choice questions from CFA practice exams and the MMLU business ethics, microeconomics and professional accounting subjects, without questions that need numerical reasoning. Zero-shot accuracy in percent, taking the option with the highest log-probability (higher is better).
Metric: Accuracy (%; 877 multiple-choice business and finance knowledge questions from CFA practice exams and three MMLU subjects, zero-shot, highest answer log-probability). Source: arxiv.org. Saturation forecast: Estimated already saturated. 16 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-4 | 80.1 | #996 |
| 2 | GPT-3.5 | 60.3 | #2430 |
| 3 | Mixtral 8x7B | 47.1 | #4051 |
| 4 | Llama 2 70B | 44.9 | #4021 |
| 5 | Mistral 7B | 44.4 | #7784 |
| 6 | falcon-40B | 43.9 | #8304 |
| 7 | mpt-30B | 42.4 | #8820 |
| 8 | Llama 2 13B | 42.1 | #8311 |
| 9 | Llama 2 7B | 41.7 | #9031 |
| 10 | falcon-7B | 40.9 | #9713 |
| 11 | mpt-7B | 39.7 | #9926 |
| 12 | starcoder | 37.9 | #9634 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=bizbench-finknow · How It Works · Data refreshed daily, snapshot 2026-11-12.