Vals Legal Index v1 (Vals Legal Index)
We show this table for reference; we do not rank on it.
Vals AI composite of agentic legal work: legal research, drafting and reviewing documents, and tax law, with practice areas weighted by their share of U.S. law-office revenue.
Evaluation results published by Vals AI. Published leaderboard aggregates and evaluation settings only. Dataset contents are not included.
Revenue-weighted accuracy on Vals Legal Index — October 27, 2026
We mirror the published revenue-weighted accuracy view for Vals Legal Index. Muse Wonder Full leads the public snapshot at 77.86%, followed by Gemini 4 Argon (76.72%) and Gemini 4 Carbon (75.83%). We do not use these results to rank models overall.
Muse Wonder Full
Meta
max reasoning
Gemini 4 Argon
high reasoning
Gemini 4 Carbon
high reasoning
30 modelsAgenticCurrentDisplay onlyUpdated October 27, 2026
Revenue-weighted accuracy table (30 models)
| Rank | Model / configuration | Score | Parameters (B) | Open / closed |
|---|---|---|---|---|
| 1 | Muse Wonder FullMetamax reasoning | 77.86% | Not reported | Closed |
| 2 | Gemini 4 ArgonGooglehigh reasoning | 76.72% | Not reported | Closed |
| 3 | Gemini 4 CarbonGooglehigh reasoning | 75.83% | Not reported | Closed |
| 4 | Muse WonderMetamax reasoning | 75.26% | Not reported | Closed |
| 5 | GLM-5.3Z.AImax reasoning | 74.58% | Not reported | Open |
| 6 | Qwen3.8 Max InternalAlibaba | 73.90% | Not reported | Unknown |
| 7 | Grok 4.7xAIxhigh reasoning | 72.60% | Not reported | Closed |
| 8 | Claude Opus 5.5Anthropic | 72.33% | Not reported | Closed |
| 9 | MiMo-V2.6-ProXiaomi | 72.14% | Not reported | Open |
| 10 | DeepSeek V4.1 Flash 1017DeepSeekmax reasoning | 71.86% | Not reported | Open |
| 11 | Kimi K3Moonshot AImax reasoning | 71.72% | Not reported | Open |
| 12 | Hy4 previewTencent | 71.70% | Not reported | Open |
| 13 | GLM-5.3-FlashZ.AImax reasoning | 71.05% | Not reported | Open |
| 14 | Claude Haiku 5.5Anthropic | 70.29% | Not reported | Closed |
| 15 | Gemini 3.8 FlashGooglehigh reasoning | 69.78% | Not reported | Closed |
| 16 | DeepSeek V4 Pro 0813DeepSeekmax reasoning | 69.01% | Not reported | Open |
| 17 | MiMo-V2.6-FlashXiaomi | 68.99% | Not reported | Open |
| 18 | GPT-6.1 SolOpenAImax reasoning | 68.72% | Not reported | Closed |
| 19 | Muse Spark OpenMetaxhigh reasoning | 68.51% | Not reported | Open |
| 20 | GPT-6 AstraOpenAImax reasoning | 68.39% | Not reported | Closed |
| 21 | Claude Sonnet 5.5Anthropic | 68.02% | Not reported | Closed |
| 22 | MiniMax M3.1 FlashMiniMax | 67.07% | Not reported | Closed |
| 23 | Kimi K3.1 ProMoonshot AImax reasoning | 67.04% | Not reported | Pending |
| 24 | Mistral Large 4Mistralhigh reasoning | 65.61% | Not reported | Open |
| 25 | Mercury 3Inceptionhigh reasoning | 62.96% | Not reported | Closed |
| 26 | GPT-6 LunaOpenAImax reasoning | 62.20% | Not reported | Closed |
| 27 | Beam 1.0Reflection | 61.80% | Not reported | Unknown |
| 28 | GPT-6.1 TerraOpenAImax reasoning | 58.64% | Not reported | Closed |
| 29 | Inkling-SmallThinking Machines Lab0.99 reasoning | 57.83% | Not reported | Open |
| 30 | InklingThinking Machines Lab0.99 reasoning | 54.16% | Not reported | Open |
How to read this leaderboard
Operator receipt: 30 sourced rows are currently displayable on this page; the leading published row is Muse Wonder Full at 77.86%.
Honest limit: It averages three Vals benchmarks with private task sets, and a model is listed only when all three have a result. Some practice areas are scored by one benchmark only (real estate, employment and estates on drafting, family law on research), and about 7% of law-office revenue has no benchmark. It measures model-and-agent runs on legal tasks, not legal practice.
How Vals Legal Index is shown here
We mirror Vals AI's Legal Index v1, updated October 24, 2026. It scores agentic legal work across eight practice areas plus tax: Legal Research Bench covers research, HLAB-Vals-1.1 covers drafting and reviewing documents, and Tax Agent Bench covers tax law.
Vals weights each practice area by its share of U.S. law-office revenue in the Census Bureau's 2022 Economic Census, so business and commercial work counts for about 40% of the score. A model is listed once all three components have a result. The results stay display only and do not enter model rankings.
Snapshot
The published Vals Legal Index snapshot places Muse Wonder Full first at 77.86%. The third row is 2.03 points behind. The broader top-10 range is 6.00 points, so many of the published results sit in a relatively narrow band.
30 models have been evaluated on Vals Legal Index. 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. Vals Legal Index is currently displayed for reference but excluded from the scoring formula, so it does not directly affect overall rankings.
About Vals Legal Index
Year
2026
Tasks
3 legal benchmarks across 8 practice areas plus tax
Format
Revenue-weighted accuracy
Difficulty
Agentic legal research, document drafting and review, and tax research
Vals published v1 on October 24, 2026. Legal Research Bench (all-pass accuracy by area of law) and HLAB-Vals-1.1 (criteria pass rate by practice area) score eight practice areas, averaged where both cover an area; the areas combine by their share of 2022 law-office revenue from the Census Bureau's Economic Census. Tax Agent Bench carries the taxation sub-line: Legal_Index = 0.978 x practice areas + 0.022 x Tax Agent Bench. We mirror the published aggregates as a display-only composite, outside overall and category rankings.
Freshness and provenance
Version
Vals Legal Index 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 Vals Legal Index measure?
Vals AI composite of agentic legal work: legal research, drafting and reviewing documents, and tax law, with practice areas weighted by their share of U.S. law-office revenue.
Which model leads the published Vals Legal Index snapshot?
Muse Wonder Full currently leads the published Vals Legal Index snapshot with 77.86% revenue-weighted accuracy. BenchLM shows this benchmark for display only and does not use it in overall rankings.
How many models are evaluated on Vals Legal Index?
The October 27, 2026 snapshot contains 30 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.