Skip to content

Benchmarks · Vals AI

Vals Legal Index

Composite index of legal research and drafting by practice area (business, civil, employment, family, criminal, tax, Legal Research Bench, HLAB), weighted by each area's share of US law-office revenue.

Models measured
36
Top
77.86 %Wonder Fullwithin the standard error of ±1.82
Version
1
Measured
9 November 2026
Go to the sourceThis benchmark does not count towards the PrudentScore.

01

Ranking

At the top: Top 15

Vals AI Legal Research Bench, all-pass [1]. Hatched bars are rejected by the compliance gate.

  1. 1Wonder Full77.86%
  2. 2Gemini 4 Argon76.78%
  3. 3Gemini 4 Carbon75.93%
  4. 4Muse Ember75.69%
  5. 5Wonder75.26%
  6. 6GLM 5.374.62%
  7. 7Inkling 2 Small73.89%
  8. 8Grok 4.772.67%
  9. 9Claude Opus 5.572.37%
  10. 10MiMo V2.6 Pro72.14%
  11. 11DeepSeek V4.1 Flash 101771.86%
  12. 12Kimi K371.72%
  13. 13Hy4 Preview71.70%
  14. 14GLM 5.3 Flash71.05%
  15. 15Hy470.55%

Source: Vals AI (Vals Legal Index) · Licence: CC BY 4.0 · Adaptation: Prudai selected the models, translated the explanations, added the gate verdict and the PrudentScore and rounded figures to two decimals.

Passes the compliance gateRejected by the compliance gate
Vals Legal Index 1
# ModelScoreCost/taskGatePrudentScore
1Wonder FullMeta77.86%± 1.82$3.72Rejectedno score
2Gemini 4 ArgonGoogle76.78%± 1.84$6.82Passes79
3Gemini 4 CarbonGoogle75.93%± 1.85$7.20Passes77
4Muse EmberMeta75.69%± 1.84$0.07Rejectedno score
5WonderMeta75.26%± 1.83$4.32Rejectedno score
6GLM 5.3Z.ai74.62%± 1.84$2.68Rejectedno score
7Inkling 2 SmallThinking Machines73.89%± 1.83$0.56Rejectedno score
8Grok 4.7xAI (SpaceXAI)72.67%± 1.82$5.82Rejectedno score
9Claude Opus 5.5Anthropic72.37%± 1.84$19.94Passes63
10MiMo V2.6 ProXiaomi72.14%± 1.83$0.18Rejectedno score
11DeepSeek V4.1 Flash 1017DeepSeek71.86%± 1.82$0.59Rejectedno score
12Kimi K3Moonshot AI71.72%± 1.80$3.47Rejectedno score
13Hy4 PreviewTencent71.70%± 1.84$0.81Rejectedno score
14GLM 5.3 FlashZ.ai71.05%± 1.83$0.23Rejectedno score
15Hy4Tencent70.55%± 1.81$1.65Rejectedno score
16Claude Haiku 5.5Anthropic70.29%± 1.83$2.36Passes59
17Step 5 PreviewStepFun70.16%± 1.78$0.84Rejectedno score
18Gemini 3.8 FlashGoogle69.78%± 1.78$2.05Passes62
19Claude Fable 5.2Anthropic69.17%± 1.81$8.88Rejectedno score
20DeepSeek V4 Pro (0813)DeepSeek69.01%± 1.77$0.87Rejectedno score
21MiMo V2.6 FlashXiaomi68.99%± 1.77$0.07Rejectedno score
22GPT-6.1 SolOpenAI68.72%± 1.77$2.75Passes59
23Muse Spark OpenMeta68.51%± 1.76$5.36Rejectedno score
24GPT-6 AstraOpenAI68.39%± 1.74$13.94Passes56
25Claude Sonnet 5.5Anthropic68.02%± 1.81$13.37Passes62
26Mistral Large 4 (GA, 26.10)Mistral AI67.84%± 1.70$0.78Not yet assessedno score
27MiniMax-M3.1-FlashMiniMax67.07%± 1.78$0.41Rejectedno score
28Kimi K3.1 ProMoonshot AI67.04%± 1.65$6.63Rejectedno score
29Mistral Large 4 (preview)Mistral AI65.61%± 1.63$2.22Rejectedno score
30Mercury 3Inception Labs62.96%± 1.62$0.10Not yet assessedno score
31GPT-6 LunaOpenAI62.20%± 1.57$0.31Passes54
32Beam 1.0Reflection AI61.80%± 1.61-Rejectedno score
33GPT-6.1 TerraOpenAI58.76%± 1.36$2.29Passes44
34Inkling SmallThinking Machines57.83%± 1.70$0.14Rejectedno score
35Nemotron 3.5 Super VLNVIDIA55.37%± 1.25$0.16Rejectedno score
36InklingThinking Machines54.16%± 1.66$0.70Rejectedno score

Source: Vals AI (Vals Legal Index) · Licence: CC BY 4.0 · Adaptation: Prudai selected the models, translated the explanations, added the gate verdict and the PrudentScore and rounded figures to two decimals.

Vals AI · Version 1 · Measured 9 November 2026 · Go to the source. The PrudentScore is a composition by Prudai B.V. based on publicly published figures and does not represent the view or endorsement of Vals AI, Harvey or Artificial Analysis. Figures from Artificial Analysis appear in running text only, with attribution and a link, as that party's terms require [9].

Other benchmarks

Claude and Anthropic, GPT and OpenAI, Gemini and Google, Grok and xAI, Kimi and Moonshot AI, Muse and Meta, Mistral, DeepSeek, GLM and Z.ai, Harvey, Vals AI and Artificial Analysis are trademarks of their respective owners. They are named for identification only and this implies no partnership, sponsorship or endorsement.