High-stakes legal professional reasoning. 550 expert-authored tasks across 47 US jurisdictions and 114 countries with rubric-based scoring. Created by 182 JD-holding professionals.
Expert-authored by lawyers; multi-jurisdictional coverage; tests realistic multi-turn legal workflows.
Top models score only ~0.37 on hard subset; heavily US-centric despite international coverage.
Higher is better. Rubric-weighted accuracy on professional legal tasks.
Top models score only ~0.37 on hard subset.
| # | Model | Score |
|---|---|---|
| 1 | 61.6 | |
| 2 | 59.8 | |
| 3 | 57.0 | |
| 4 | 53.4 | |
| 5 | 52.6 | |
| 6 | 52.3 | |
| 7 | 51.6 | |
| 8 | 50.5 | |
| 9 | 49.9 | |
| 10 | 49.7 | |
| 11 | 49.3 | |
| 12 | 49.0 |