Skip to content

Benchmarks · Vals AI

Vals Index

General agent index across finance, coding, legal and tax, weighted by each sector's share of US GDP. Version 2.2 is current; 2.1 is archived.

Models measured
50
Top
68.90 %Gemini 4 Argonwithin the standard error of ±0.97
Version
2.1
Measured
18 October 2026
Go to the sourceThis benchmark does not count towards the PrudentScore.
Versions

01

Ranking

At the top: Top 15

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

  1. 1Gemini 4 Argon68.90%
  2. 2Claude Sonnet 5.567.04%
  3. 3Claude Opus 5.566.97%
  4. 4Claude Fable 5.165.83%
  5. 5Wonder Full64.57%
  6. 6Claude Opus 563.67%
  7. 7GPT-6 Astra63.13%
  8. 8Claude Fable 561.39%
  9. 9GPT-6.1 Sol61.15%
  10. 10Wonder61.06%
  11. 11Muse Spark 1.3 (Max)58.16%
  12. 12GPT-5.6 Sol58.01%
  13. 13GPT-6 Sol57.54%
  14. 14DeepSeek V4.1 Flash 101756.58%
  15. 15MiMo V2.6 Pro55.20%

Source: Vals AI (Vals 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 Index 2.1
# ModelScoreCost/taskGatePrudentScore
1Gemini 4 ArgonGoogle68.90%± 0.97$15.68Passes79
2Claude Sonnet 5.5Anthropic67.04%± 0.92$21.34Passes62
3Claude Opus 5.5Anthropic66.97%± 0.89$32.14Passes63
4Claude Fable 5.1Anthropic65.83%± 1.11$28.71Rejectedno score
5Wonder FullMeta64.57%± 1.07$8.56Rejectedno score
6Claude Opus 5Anthropic63.67%± 0.96$19.31Passes72
7GPT-6 AstraOpenAI63.13%± 1.19$18.46Passes56
8Claude Fable 5Anthropic61.39%± 1.00$29.59Rejectedno score
9GPT-6.1 SolOpenAI61.15%± 1.01$3.24Passes59
10WonderMeta61.06%± 1.07$11.44Rejectedno score
11Muse Spark 1.3 (Max)Meta58.16%± 1.19$3.79Rejectedno score
12GPT-5.6 SolOpenAI58.01%± 1.02$14.24Passes57
13GPT-6 SolOpenAI57.54%± 1.01$7.58Passes45
14DeepSeek V4.1 Flash 1017DeepSeek56.58%± 1.27$0.64Rejectedno score
15MiMo V2.6 ProXiaomi55.20%± 1.18$0.41Rejectedno score
16Claude Opus 4.8Anthropic55.10%± 1.01$13.14Passes62
17Grok 4.7xAI (SpaceXAI)54.95%± 1.07$12.11Rejectedno score
18Gemini 3.8 FlashGoogle54.89%± 1.02$5.71Passes62
19Claude Haiku 5.5Anthropic54.44%± 1.41$3.00Passes59
20GLM 5.3Z.ai53.51%± 1.30$7.25Rejectedno score
21MiMo V2.6 FlashXiaomi53.23%± 1.10$0.20Rejectedno score
22Muse Spark 1.3Meta53.20%± 1.07$3.38Rejectedno score
23GPT-5.6 TerraOpenAI53.09%± 1.29$5.78Passes53
24Grok 4.6xAI (SpaceXAI)52.09%± 1.15$4.49Rejectedno score
25Claude Sonnet 5Anthropic51.90%± 1.10$13.69Passes59
26GPT-5.6 LunaOpenAI51.69%± 1.06$0.82Passes56
27DeepSeek V4.1 FlashDeepSeek51.32%± 1.13$0.33Rejectedno score
28Gemini 3.7 FlashGoogle51.27%± 1.10$4.67Passes58
29GPT-6 LunaOpenAI51.22%± 1.07$0.43Passes54
30Ember-1Fireworks AI50.94%± 1.30$9.66Not yet assessedno score
31Kimi K3Moonshot AI50.30%± 0.99$6.44Rejectedno score
32Hy4 PreviewTencent49.94%± 1.15$1.44Rejectedno score
33Step 5 PreviewStepFun49.35%± 1.22$2.63Rejectedno score
34Muse Spark 1.2Meta49.29%± 1.10$1.94Rejectedno score
35Mistral Large 4 (preview)Mistral AI48.66%± 1.13$14.07Rejectedno score
36Qwen 3.8 MaxAlibaba48.27%± 1.34$4.96Not yet assessedno score
37DeepSeek V4 Flash 0731DeepSeek47.96%± 1.24$0.85Rejectedno score
38DeepSeek V4 Pro (0813)DeepSeek47.63%± 1.12$3.35Rejectedno score
39Muse Spark OpenMeta46.61%± 1.20$4.08Rejectedno score
40Gemini 3.5 FlashGoogle44.79%± 1.10$3.30Passes47
41Grok 4.5xAI (SpaceXAI)44.72%± 1.17$2.42Rejectedno score
42GLM 5.3 FlashZ.ai44.58%± 1.41$1.03Rejectedno score
43DeepSeek V4 Pro (2026-04-23)DeepSeek38.63%± 1.16$1.31Rejectedno score
44MiniMax-M3MiniMax36.53%± 1.17$3.08Rejectedno score
45MiMo V2.5 ProXiaomi33.84%± 1.17$0.16Rejectedno score
46Gemini 3.1 Pro Preview (02/26)Google33.44%± 1.13$2.18Rejectedno score
47GPT 5.4 MiniOpenAI33.17%± 1.28$1.45Passes25
48InklingThinking Machines28.67%± 1.10$1.46Rejectedno score
49Inkling SmallThinking Machines25.46%± 1.03$0.28Rejectedno score
50Mercury 2.5Inception Labs8.86%± 0.45$0.29Not yet assessedno score

Source: Vals AI (Vals 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 2.1 · Measured 18 October 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.