LegalBench measures legal reasoning across tasks such as rule recall, interpretation, argumentation, and rhetoric. It is useful for comparing legal-assistant potential before domain-specific validation.
Large multi-task suite provides broad coverage of legal reasoning behaviors; regularly updated leaderboard supports current frontier comparisons.
Task design is primarily US-centric, so transfer to other jurisdictions is limited; some subtasks reward statute recall more than practical legal judgment.
Higher is better. Accuracy percentage across all 162 tasks.
Max score: 100
Task design is primarily US-centric, so transfer to other jurisdictions is limited.
| # | Model | Score |
|---|---|---|
| 1 | 97.0 | |
| 2 | 96.9 | |
| 3 | 96.1 | |
| 4 | 94.3 | |
| 5 | 94.3 | |
| 6 | 93.8 | |
| 7 | 93.5 | |
| 8 | 93.4 | |
| 9 | 93.2 | |
| 10 | 93.2 | |
| 11 | 93.2 | |
| 12 | 92.9 | |
| 13 | 92.9 | |
| 14 | 92.7 | |
| 15 | 92.7 | |
| 16 | 92.7 | |
| 17 | 92.6 | |
| 18 | 92.3 | |
| 19 | 92.2 | |
| 20 | 92.0 | |
| 21 | 91.9 | |
| 22 | 91.6 | |
| 23 | 91.5 | |
| 24 | 91.4 | |
| 25 | 91.3 | |
| 26 | 91.2 | |
| 27 | 91.1 | |
| 28 | 91.0 | |
| 29 | 91.0 | |
| 30 | 90.8 | |
| 31 | 90.7 | |
| 32 | 90.6 | |
| 33 | 90.4 | |
| 34 | 90.4 | |
| 35 | 90.3 | |
| 36 | 90.3 | |
| 37 | 90.2 | |
| 38 | 90.2 | |
| 39 | 90.2 | |
| 40 | 90.2 | |
| 41 | 90.2 | |
| 42 | 90.1 | |
| 43 | 90.1 | |
| 44 | 90.1 | |
| 45 | 90.1 | |
| 46 | 90.0 | |
| 47 | 89.9 | |
| 48 | 89.8 | |
| 49 | 89.8 | |
| 50 | 89.8 |