Benchmarks › Vals Legal Research Bench
55.3top score · Claude Fable 5.1
69models ranked
28open-weights models
All-pass accuracy; secondary weighted partial credithigher is better
Oct 20, 2026last updated
How the test works
1
Task
Research a legal question using search and document tools
2
Run
Yes
- Tools allowedCase law search, web search, document retrieval
3
Score
All-pass accuracy; secondary weighted partial credit
- Graded byLLM judge on lawyer-written required rubric items
- Better isHigher
Sources: vals.ai
Vals Legal Research Bench leaderboardtop 10 of 69
Full leaderboard with filters →| Model | Vals Legal Research Bench | Per task | Price | Speed | Supports | Context |
|---|---|---|---|---|---|---|
55.3unstated | — | $10 / $50 | — | 1M | ||
55.3unstated | — | $5 / $25 | — | 1M | ||
55.3max | $1.60 | $1.25 / $4.25 | 128 | 1M | ||
54.8unstated | — | $2 / $10 | — | 1M | ||
50.5unstated | — | $4 / $20 | — | 1M | ||
49.5unstated | — | $10 / $50 | — | 1M | ||
49unstated | — | $1.40 / $4.40 | — | 1M | ||
48.1unstated | — | $2 / $10 | — | 1M | ||
48.1unstated | — | $4 / $20 | — | 1.05M | ||
48.1unstated | — | $2 / $6 | — | 500K |
Vals Legal Research Bench questions
What is Vals Legal Research Bench?›
Research a legal question using search and document tools. Created by Vals AI.
Which model is best on Vals Legal Research Bench?›
Claude Fable 5.1 leads Vals Legal Research Bench with 55.3, ahead of Claude Opus 5 (55.3) and Muse Spark 1.3 (55.3).
What is the best open-weights model on Vals Legal Research Bench?›
GLM 5.3 with 49, then MiMo v2.6 Pro (47.1).
How is Vals Legal Research Bench scored?›
All-pass accuracy; secondary weighted partial credit; llm judge on lawyer-written required rubric items.