Mistral 7B: benchmark results
Mistral AI's original 7B base model, a transformer with grouped-query and sliding-window attention, released under Apache 2.0 in September 2023. Provider: Mistral. Released 2023-09-27. Access: Open.
Aggregate Rating 1343 ± 11, rank #1397 of 1656 rated models, from 73 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| LegalCiteBench - Case Verification | 96.07 | Cat4-2 judge score (0-100) for verifying whether a cited cas | 100 |
| The Wrong Kind of Right | 5.7 | Misfired Alignment Rate, average over 8 categories (%) (self | 79.2 |
| BizBench - ConvFinQA Extract | 91.5 | Accuracy (%; 916 ConvFinQA questions answerable by extractin | 66.7 |
| BizBench - FinKnow | 44.4 | Accuracy (%; 877 multiple-choice business and finance knowle | 66.7 |
| BizBench - TAT-QA Extract | 87.9 | Accuracy (%; 248 TAT-QA questions answerable by extracting a | 66.7 |
| LitBench - Citation Recommendation (Quantitative Biology) | 27.94 | Accuracy (%) at picking the paper actually cited, out of 11 | 66.7 |
| LitBench - Citation Recommendation (Robotics) | 18.8 | Accuracy (%) at picking the paper actually cited, out of 11 | 66.7 |
| BizBench | 61.8 | Mean accuracy (%) over the eight BizBench tasks (FinKnow, Fo | 64.3 |
| BizBench - CodeFinQA | 48.8 | Accuracy (%; 844 FinQA questions over report text and tables | 60 |
| BizBench - FinCode | 15.7 | Accuracy (%; 121 CFA and CPA exam questions answered by gene | 60 |
| BizBench - FormulaEval | 66 | Accuracy (%; 50 financial-formula functions to complete from | 60 |
| BizBench - SEC-Num | 65.1 | Accuracy (%; 2,000 SEC 10-K and 10-Q filing snippets, extrac | 60 |
Interactive version: theaggregate.ai/model?slug=mistral-7b · How It Works · Data refreshed daily, snapshot 2026-10-20.