Skip to main content

Benchmark profile

Vals Legal Research Bench (Legal Research Bench)

Evaluating agents on legal research tasks across diverse areas of US law

Data verified

How BenchLM shows Legal Research Bench

BenchLM mirrors the public Vals AI Legal Research Bench leaderboard captured from https://www.vals.ai/benchmarks/legal_research and updated by Vals on July 18, 2026. The snapshot preserves overall scores, uncertainty, latency, cost-per-test metadata, and task-level scores where Vals publishes them.

Legal Research Bench is display only on BenchLM. Vals proprietary or Vals-hosted aggregate views are useful context, but BenchLM does not use them as weighted ranking inputs or as a replacement for benchmark-native source records.

24 Vals rows18 task viewsprivate datasetTasks: Overall, Administrative / Regulatory, Business & Commercial, Civil Litigation, Constitutional / Civil RightsDisplay only

Legal Research Bench score on Legal Research Bench — July 18, 2026

BenchLM mirrors the published legal research bench score view for Legal Research Bench. Claude Fable 5 leads the public snapshot at 49.52% , followed by GPT-5.6 Sol (48.08%) and Kimi K3 (44.23%). BenchLM does not use these results to rank models overall.

24 modelsExternal benchmark mirrorsCurrentDisplay onlyUpdated July 18, 2026

Legal Research Bench score table (24 models)

Score
1
49.52%
2
48.08%
3
Kimi K3Moonshot AI
44.23%
4
43.75%
5
41.83%
6
40.87%
7
GPT-5.5OpenAI
40.38%
8
38.46%
9
37.98%
10
Grok 4.5SpaceXAI
37.98%
11
36.54%
12
GLM 5.2Zhipu AI
31.25%
13
30.77%
14
MiniMax M3MiniMax
29.81%
15
InklingThinkingmachines
28.36%
16
GLM 5.1Zhipu AI
27.89%
17
25.48%
18
23.08%
20
Kimi K2.6Moonshot AI
15.87%
21
Grok 4.3SpaceXAI
15.38%
22
12.50%
23
Laguna M.1Poolside
2.40%
24
Laguna Xs.2Poolside
0.96%

The published Legal Research Bench snapshot places Claude Fable 5 first at 49.52%. The third row is 5.29 points behind. The broader top-10 range is 11.54 points, so the table still separates the published systems.

24 models have been evaluated on Legal Research Bench. The benchmark falls in the External benchmark mirrors category. We keep external benchmark mirrors separate from the weighted global scoring system, so these results remain source-specific evidence. Legal Research Bench is currently displayed for reference but excluded from the scoring formula, so it does not directly affect overall rankings.

About Legal Research Bench

Year

2026

Tasks

US-law legal research tasks

Format

Accuracy score

Difficulty

Professional legal research

BenchLM mirrors the public Vals AI Legal Research Bench leaderboard as display-only external evidence. The captured snapshot preserves overall scores, task-level scores where Vals publishes them, uncertainty, latency, and cost-per-test metadata. It is excluded from BenchLM weighted rankings.

BenchLM freshness & provenance

Version

Legal Research Bench 2026

Refresh cadence

Quarterly

Staleness state

Current

Question availability

Public benchmark set

CurrentDisplay only

BenchLM uses freshness metadata to decide whether a benchmark should still be treated as a strong differentiator, a benchmark to watch, or a display-only reference. For the full scoring policy, see the BenchLM methodology page.

FAQ

What does Legal Research Bench measure?

Evaluating agents on legal research tasks across diverse areas of US law

Which model leads the published Legal Research Bench snapshot?

Claude Fable 5 currently leads the published Legal Research Bench snapshot with 49.52% legal research bench score. BenchLM shows this benchmark for display only and does not use it in overall rankings.

How many models are evaluated on Legal Research Bench?

24 AI models are included in BenchLM's mirrored Legal Research Bench snapshot, based on the public leaderboard captured on July 18, 2026.

Last updated: July 18, 2026 · mirrored from the public benchmark leaderboard

Choose a model with this week’s evidence

Join 2,000+ readers for ranking moves, pricing changes, and the claims that still need proof.

One email each week. Unsubscribe anytime.