Benchmarks › Vals LegalBench
88.6top score · Claude Fable 5
89models ranked
41open-weights models
Accuracyhigher is better
Oct 20, 2026last updated
How the test works
1
Task
Answer legal reasoning questions
- Kinds of tasksIssue spotting, rule recall, rule conclusion, rule application, interpretation, rhetorical understanding
- ExampleWhere in the Federal Rules of Civil Procedure are notice requirements described?
2
Score
Accuracy
- Graded byVals correctness checks; exact LegalBench checker breakdown not published
- Better isHigher
Sources: vals.ai
Vals LegalBench leaderboardtop 10 of 89
Full leaderboard with filters →| Model | Vals LegalBench | Per task | Price | Speed | Supports | Context |
|---|---|---|---|---|---|---|
88.6unstated | — | $10 / $50 | — | 1M | ||
88.5unstated | — | $10 / $50 | — | 1M | ||
88.3unstated | — | $2 / $10 | — | 1M | ||
87.4unstated | — | $2 / $12 | — | 1M | ||
87unstated | — | $0.75 / $3.75 | — | 1M | ||
87unstated | — | $5 / $25 | — | 1M | ||
87unstated | — | $4 / $20 | — | 1.05M | ||
86.9unstated | — | $0.50 / $3 | — | 1M | ||
86.7unstated | — | $0.75 / $3.75 | — | 1M | ||
86.5unstated | — | $5 / $30 | — | 1.05M |
Vals LegalBench questions
What is Vals LegalBench?›
Answer legal reasoning questions. Created by Vals AI (LegalBench evaluation).
Which model is best on Vals LegalBench?›
Claude Fable 5 leads Vals LegalBench with 88.6, ahead of Claude Fable 5.1 (88.5) and Gemini 4 Argon (88.3).
What is the best open-weights model on Vals LegalBench?›
Kimi K3 with 86.2, then MiniMax M3 (85.4).
How is Vals LegalBench scored?›
Accuracy; vals correctness checks; exact legalbench checker breakdown not published.