Vals Legal Research Bench (Legal Research Bench)
We show this table for reference; we do not rank on it.
Evaluating agents on legal research tasks across diverse areas of US law
Evaluation results published by Vals AI. Published leaderboard aggregates and evaluation settings only. Dataset contents are not included.
Legal Research Bench score on Legal Research Bench — October 24, 2026
We mirror the published legal research bench score view for Legal Research Bench. Muse Spark 1.3 Max leads the public snapshot at 55.29%, followed by Claude Opus 5 (55.29%) and Claude Fable 5.1 (55.29%). We do not use these results to rank models overall.
Muse Spark 1.3 Max
Meta
max reasoning
Claude Opus 5
Anthropic
Claude Fable 5.1
Anthropic
85 modelsAgenticCurrentDisplay onlyUpdated October 24, 2026
Legal Research Bench score table (85 models)
| Rank | Model / configuration | Score | Parameters (B) | Open / closed |
|---|---|---|---|---|
| 1 | Muse Spark 1.3 MaxMetamax reasoning | 55.29% | Not reported | Unknown |
| 2 | Claude Opus 5Anthropic | 55.29% | Not reported | Closed |
| 3 | Claude Fable 5.1Anthropic | 55.29% | Not reported | Closed |
| 4 | Gemini 4 ArgonGooglehigh reasoning | 54.81% | Not reported | Closed |
| 5 | Muse Wonder FullMetamax reasoning | 53.37% | Not reported | Closed |
| 6 | Muse WonderMetamax reasoning | 50.48% | Not reported | Closed |
| 7 | Claude Opus 5.5Anthropic | 50.48% | Not reported | Closed |
| 8 | Claude Fable 5Anthropic | 49.52% | Not reported | Closed |
| 9 | GLM-5.3Z.AImax reasoning | 49.04% | Not reported | Open |
| 10 | Grok 4.6xAIhigh reasoning | 48.08% | Not reported | Closed |
| 11 | GPT-5.6 SolOpenAImax reasoning | 48.08% | Not reported | Closed |
| 12 | Claude Sonnet 5.5Anthropic | 48.08% | Not reported | Closed |
| 13 | Qwen3.8 MaxAlibaba | 47.60% | Not reported | Open |
| 14 | Grok 4.7xAIxhigh reasoning | 47.12% | Not reported | Closed |
| 15 | MiMo-V2.6-ProXiaomi | 47.12% | Not reported | Open |
| 16 | Kimi K3Moonshot AImax reasoning | 46.15% | Not reported | Pending |
| 17 | Qwen3.8 Max InternalAlibaba | 46.15% | Not reported | Unknown |
| 18 | DeepSeek V4.1 Flash 1017DeepSeekmax reasoning | 45.19% | Not reported | Open |
| 19 | GLM-5.3-FlashZ.AImax reasoning | 45.19% | Not reported | Open |
| 20 | Hy4 previewTencent | 45.19% | Not reported | Open |
| 21 | Ember-1Fireworks | 44.71% | Not reported | Closed |
| 22 | Muse Spark 1.2Metaxhigh reasoning | 43.75% | Not reported | Closed |
| 23 | Claude Opus 4.8Anthropic | 43.75% | Not reported | Closed |
| 24 | Claude Haiku 5.5Anthropic | 43.27% | Not reported | Closed |
| 25 | Claude Sonnet 5Anthropic | 41.83% | Not reported | Closed |
| 26 | GLM LegalZhipu AImax reasoning | 41.83% | Not reported | Unknown |
| 27 | DeepSeek V4.1 Flash 1017DeepSeekmax reasoning | 41.35% | Not reported | Open |
| 28 | GPT-5.6 TerraOpenAImax reasoning | 41.35% | Not reported | Closed |
| 29 | Muse Spark 1.3Metaxhigh reasoning | 40.87% | Not reported | Closed |
| 30 | DeepSeek V4 Pro 0813DeepSeekmax reasoning | 40.87% | Not reported | Open |
| 31 | Step 5 PreviewStepFun | 40.38% | Not reported | Open |
| 32 | GPT-5.5OpenAIxhigh reasoning | 40.38% | Not reported | Closed |
| 33 | GPT-6 AstraOpenAImax reasoning | 39.42% | Not reported | Closed |
| 34 | Gemini 3.8 FlashGooglehigh reasoning | 38.94% | Not reported | Closed |
| 35 | Claude Opus 4.7Anthropic | 38.46% | Not reported | Closed |
| 36 | Claude Sonnet 4.6Anthropic | 38.46% | Not reported | Closed |
| 37 | Muse Spark OpenMetaxhigh reasoning | 38.46% | Not reported | Open |
| 38 | GPT-6.1 SolOpenAImax reasoning | 38.46% | Not reported | Closed |
| 39 | Muse Spark 1.1Metaxhigh reasoning | 37.98% | Not reported | Closed |
| 40 | MiMo-V2.6-FlashXiaomi | 37.98% | Not reported | Open |
| 41 | Grok 4.5xAIhigh reasoning | 37.98% | Not reported | Closed |
| 42 | Kimi K3.1 ProMoonshot AImax reasoning | 37.50% | Not reported | Pending |
| 43 | GPT-5.6 LunaOpenAImax reasoning | 36.54% | Not reported | Closed |
| 44 | Qwen3.8-27BAlibabaxhigh reasoning | 36.06% | Not reported | Open |
| 45 | Gemini 3.7 FlashGooglehigh reasoning | 34.62% | Not reported | Closed |
| 46 | MiniMax M3.1 FlashMiniMax | 34.13% | Not reported | Closed |
| 47 | Mistral Large 4Mistralhigh reasoning | 31.73% | Not reported | Pending |
| 48 | GLM-5.2Z.AImax reasoning | 31.25% | Not reported | Open |
| 49 | Gemini 3.5 FlashGooglehigh reasoning | 30.77% | Not reported | Closed |
| 50 | DeepSeek V4 Flash 0731DeepSeekhigh reasoning | 30.29% | Not reported | Open |
| 51 | GPT-6 LunaOpenAImax reasoning | 30.29% | Not reported | Closed |
| 52 | MiniMax M3MiniMax | 29.81% | Not reported | Open |
| 53 | Beam 1.0Reflection | 28.85% | Not reported | Unknown |
| 54 | Mercury 3Inceptionhigh reasoning | 28.85% | Not reported | Closed |
| 55 | GPT-6 SolOpenAImax reasoning | 28.85% | Not reported | Closed |
| 56 | InklingThinking Machines Lab0.99 reasoning | 28.36% | Not reported | Open |
| 57 | GLM-5.1Z.AI | 27.89% | Not reported | Open |
| 58 | Inkling-SmallThinking Machines Lab0.99 reasoning | 25.48% | Not reported | Open |
| 59 | Qwen3.7 MaxAlibaba | 25.48% | Not reported | Closed |
| 60 | Gemini 3.6 FlashGooglehigh reasoning | 25.00% | Not reported | Closed |
| 61 | DeepSeek V4 Pro 0813DeepSeekmax reasoning | 23.08% | Not reported | Open |
| 62 | Gemini 3.1 Pro PreviewGooglehigh reasoning | 20.67% | Not reported | Unknown |
| 63 | Gemini 3 Flash PreviewGooglehigh reasoning | 18.27% | Not reported | Unknown |
| 64 | Qwen3.7 PlusAlibaba | 16.35% | Not reported | Closed |
| 65 | MiMo-V2.5-ProXiaomi | 15.87% | Not reported | Closed |
| 66 | Kimi K2.5 ThinkingMoonshot AI | 15.87% | Not reported | Unknown |
| 67 | Kimi K2.6Moonshot AI | 15.87% | Not reported | Open |
| 68 | Grok 4.3xAIhigh reasoning | 15.38% | Not reported | Closed |
| 69 | Ling 3.0 Flash Af Rc3Ant | 15.38% | Not reported | Unknown |
| 70 | Nemotron 3 Ultra 550b A55bNvidia | 15.38% | Not reported | Unknown |
| 71 | Qwen3.6 PlusAlibaba | 14.90% | Not reported | Closed |
| 72 | Gemini 3.5 Flash-LiteGooglehigh reasoning | 13.94% | Not reported | Closed |
| 73 | Grok 4.20 0309 ReasoningxAI | 13.94% | Not reported | Unknown |
| 74 | GPT-5.4 miniOpenAIxhigh reasoning | 12.50% | Not reported | Closed |
| 75 | Claude Haiku 4.5 ThinkingAnthropic | 10.58% | Not reported | Closed |
| 76 | MiniMax M2.7MiniMax | 10.58% | Not reported | Open |
| 77 | MiMo-V2.5Xiaomi | 9.13% | Not reported | Closed |
| 78 | Mistral Medium 3.5Mistral AIhigh reasoning | 9.13% | Not reported | Unknown |
| 79 | GPT-5.4 nanoOpenAIhigh reasoning | 6.25% | Not reported | Closed |
| 80 | Mercury 2.5Inceptionhigh reasoning | 4.33% | Not reported | Closed |
| 81 | Gemini 3.1 Flash Lite PreviewGooglehigh reasoning | 3.37% | Not reported | Unknown |
| 82 | Nemotron Lightning 3p5 30b A3bFireworks AI | 2.88% | Not reported | Unknown |
| 83 | Laguna M.1Poolside | 2.40% | Not reported | Closed |
| 84 | Laguna XS.2Poolside | 0.96% | Not reported | Open |
| 85 | Ling 3.0 Flash 2607Ant | 0.00% | Not reported | Unknown |
How Legal Research Bench is shown here
BenchLM mirrors the public Vals AI Legal Research Bench leaderboard captured from https://www.vals.ai/benchmarks/legal_research and updated by Vals on October 24, 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.
Snapshot
The published Legal Research Bench snapshot places Muse Spark 1.3 Max first at 55.29%. The third row is 0.00 points behind. The broader top-10 range is 7.21 points, so many of the published results sit in a relatively narrow band.
85 models have been evaluated on Legal Research Bench. The benchmark falls in the Agentic 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.
Freshness and provenance
Version
Legal Research Bench 2026
Refresh cadence
Quarterly
Staleness state
Current
Question availability
Public benchmark set
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.
Questions
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?
Muse Spark 1.3 Max currently leads the published Legal Research Bench snapshot with 55.29% 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?
The October 24, 2026 snapshot contains 85 AI models.
Know when it’s worth switching models
The model to choose, the cheaper alternative, and the release we would wait on.
Read a sample issueJoin 6,100+ subscribers.
One email each week. Unsubscribe anytime.