Benchmarks · Vals AI
Harvey LAB (Vals implementation)
Note: Vals has no benchmark aimed at Dutch or EU law. Measures whether an agent performs real legal work with documents, spreadsheets, presentations and file-system tools (tasks from Harvey's Legal Agent Benchmark, 24 practice areas). "All-pass" counts only when every rubric criterion of a task is met; "criteria-pass-rate" is the share of criteria met. Version 1.1 has been the default since 19 October 2026; version 1 is archived (last update 18 October 2026) and is what the current PrudentBench page cites.
- Models measured
- 80
- Top
- 25.42 %Muse Spark 1.2within the standard error of ±3.67
- Version
- 1
- Measured
- 18 October 2026
01
Ranking
At the top: Top 15
Vals AI Legal Research Bench, all-pass [1]. Hatched bars are rejected by the compliance gate.
- 1Muse Spark 1.225.42%
- 2Muse Spark 1.3 (Max)23.75%
- 3Muse Spark 1.322.92%
- 4Wonder Full21.67%
- 5Muse Spark 1.120.00%
- 6Gemini 4 Argon19.58%
- 7GLM Legal19.17%
- 8Wonder17.08%
- 9Muse Spark Open16.67%
- 10=Grok 4.615.83%
- 10=Mistral Large 4 (preview)15.83%
- 12=Grok 4.512.92%
- 12=Kimi K312.92%
- 14Grok 4.712.50%
- 15=Claude Fable 511.25%
Source: Vals AI (Harvey LAB (Vals implementation)) · 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.
| # Model | Score | Cost/task | Gate | PrudentScore |
|---|---|---|---|---|
| 1Muse Spark 1.2Meta | 25.42%± 3.67 | $2.09 | Rejected | no score |
| 2Muse Spark 1.3 (Max)Meta | 23.75%± 3.55 | $2.26 | Rejected | no score |
| 3Muse Spark 1.3Meta | 22.92%± 3.42 | $2.76 | Rejected | no score |
| 4Wonder FullMeta | 21.67%± 3.54 | $5.41 | Rejected | no score |
| 5Muse Spark 1.1Meta | 20.00%± 2.94 | $0.80 | Rejected | no score |
| 6Gemini 4 ArgonGoogle | 19.58%± 3.31 | $10.55 | Passes | 79 |
| 7GLM LegalZ.ai | 19.17%± 3.25 | $0.15 | Rejected | no score |
| 8WonderMeta | 17.08%± 3.10 | $7.08 | Rejected | no score |
| 9Muse Spark OpenMeta | 16.67%± 3.09 | $6.83 | Rejected | no score |
| 10=Grok 4.6xAI (SpaceXAI) | 15.83%± 2.65 | $4.01 | Rejected | no score |
| 10=Mistral Large 4 (preview)Mistral AI | 15.83%± 2.96 | $4.18 | Rejected | no score |
| 12=Grok 4.5xAI (SpaceXAI) | 12.92%± 2.53 | $2.02 | Rejected | no score |
| 12=Kimi K3Moonshot AI | 12.92%± 2.68 | $3.83 | Rejected | no score |
| 14Grok 4.7xAI (SpaceXAI) | 12.50%± 2.66 | $11.13 | Rejected | no score |
| 15=Claude Fable 5Anthropic | 11.25%± 2.15 | $19.23 | Rejected | no score |
| 15=MiMo V2.6 FlashXiaomi | 11.25%± 2.54 | $0.09 | Rejected | no score |
| 15=Qwen 3.8 27BAlibaba | 11.25%± 2.42 | $4.03 | Rejected | no score |
| 18MiMo V2.6 ProXiaomi | 10.83%± 2.45 | $0.22 | Rejected | no score |
| 19Qwen 3.8 MaxAlibaba | 10.42%± 2.29 | $2.37 | Not yet assessed | no score |
| 20Gemini 3.8 FlashGoogle | 10.00%± 2.42 | $3.66 | Passes | 62 |
| 21Claude Opus 4.8Anthropic | 9.58%± 2.15 | $10.22 | Passes | 62 |
| 22=DeepSeek V4.1 Flash 1017DeepSeek | 9.17%± 2.51 | $0.54 | Rejected | no score |
| 22=Ember-1Fireworks AI | 9.17%± 2.44 | $3.33 | Not yet assessed | no score |
| 22=Hy4 PreviewTencent | 9.17%± 2.00 | $0.98 | Rejected | no score |
| 22=Step 5 PreviewStepFun | 9.17%± 2.44 | $1.00 | Rejected | no score |
| 26Gemini 3.7 FlashGoogle | 8.75%± 2.15 | $2.56 | Passes | 58 |
| 27=DeepSeek V4 Flash 0731DeepSeek | 8.33%± 2.15 | $0.06 | Rejected | no score |
| 27=GLM 5.3Z.ai | 8.33%± 2.00 | $4.30 | Rejected | no score |
| 29DeepSeek V4 Pro (0813)DeepSeek | 7.50%± 2.15 | $0.17 | Rejected | no score |
| 30GLM 5.2Z.ai | 7.08%± 2.00 | $2.06 | Rejected | no score |
| 31=Claude Fable 5.1Anthropic | 6.67%± 1.17 | $46.21 | Rejected | no score |
| 31=Claude Opus 4.7Anthropic | 6.67%± 2.00 | $10.81 | Passes | 54 |
| 31=Claude Opus 5Anthropic | 6.67%± 1.65 | $23.67 | Passes | 72 |
| 31=DeepSeek V4.1 FlashDeepSeek | 6.67%± 1.83 | $0.17 | Rejected | no score |
| 31=GLM 5.3 FlashZ.ai | 6.67%± 1.65 | $0.57 | Rejected | no score |
| 36=GPT-6 AstraOpenAI | 5.42%± 1.17 | $26.16 | Passes | 56 |
| 36=GPT-6.1 SolOpenAI | 5.42%± 1.75 | $3.93 | Passes | 59 |
| 38=Claude Sonnet 4.6Anthropic | 5.00%± 1.65 | $3.04 | Passes | 55 |
| 38=Claude Sonnet 5Anthropic | 5.00%± 1.65 | $8.95 | Passes | 59 |
| 40MiniMax-M3MiniMax | 4.17%± 1.65 | $1.46 | Rejected | no score |
| 41=Claude Opus 5.5Anthropic | 3.75%± 1.47 | $21.38 | Passes | 63 |
| 41=DeepSeek V4 Pro (2026-04-23)DeepSeek | 3.75%± 1.43 | $0.74 | Rejected | no score |
| 41=GPT 5.5OpenAI | 3.75%± 1.17 | $4.60 | Passes | 51 |
| 44Gemini 3.6 FlashGoogle | 3.33%± 1.17 | $1.90 | Passes | 46 |
| 45=Claude Sonnet 5.5Anthropic | 2.92%± 1.36 | $16.61 | Passes | 62 |
| 45=GPT-6 LunaOpenAI | 2.92%± 1.36 | $0.30 | Passes | 54 |
| 47=Gemini 3.5 FlashGoogle | 2.50%± 0.83 | $2.12 | Passes | 47 |
| 47=GPT-5.6 SolOpenAI | 2.50%± 0.83 | $10.37 | Passes | 57 |
| 49=InklingThinking Machines | 2.08%± 0.83 | $1.25 | Rejected | no score |
| 49=MiMo V2.5 ProXiaomi | 2.08%± 0.83 | $0.11 | Rejected | no score |
| 51=GPT-6 SolOpenAI | 1.67%± 0.82 | $3.36 | Passes | 45 |
| 51=Inkling SmallThinking Machines | 1.67%± 0.83 | $0.17 | Rejected | no score |
| 51=Kimi K2.6Moonshot AI | 1.67%± 0.83 | $0.82 | Rejected | no score |
| 51=MiMo V2.5Xiaomi | 1.67%± 0.00 | $0.04 | Rejected | no score |
| 51=Qwen 3.7 MaxAlibaba | 1.67%± 0.83 | $1.58 | Not yet assessed | no score |
| 56=Claude Haiku 5.5Anthropic | 1.25%± 0.72 | $3.39 | Passes | 59 |
| 56=GPT-5.6 LunaOpenAI | 1.25%± 0.83 | $0.38 | Passes | 56 |
| 56=Ling 3.0 FlashAnt Group | 1.25%± 0.83 | $0.05 | Rejected | no score |
| 56=Qwen 3.6 PlusAlibaba | 1.25%± 0.83 | $3.94 | Not yet assessed | no score |
| 60=Claude Haiku 4.5 (Thinking)Anthropic | 0.83%± 0.83 | $0.42 | Passes | 32 |
| 60=GPT-5.6 TerraOpenAI | 0.83%± 0.83 | $3.20 | Passes | 53 |
| 62=Grok 4.3xAI (SpaceXAI) | 0.42%± 0.00 | $0.39 | Rejected | no score |
| 62=Mistral Medium 3.5Mistral AI | 0.42%± 0.00 | $6.07 | Passes | 23 |
| 62=Nemotron 3 UltraNVIDIA | 0.42%± 0.00 | - | Rejected | no score |
| 65=Gemini 3 Flash (12/25)Google | 0.00%± 0.00 | $0.26 | Rejected | no score |
| 65=Gemini 3.1 Flash Lite PreviewGoogle | 0.00%± 0.00 | $0.08 | Not yet assessed | no score |
| 65=Gemini 3.1 Pro Preview (02/26)Google | 0.00%± 0.00 | $1.15 | Rejected | no score |
| 65=Gemini 3.5 Flash LiteGoogle | 0.00%± 0.00 | $0.33 | Passes | 38 |
| 65=GLM 5.1Z.ai | 0.00%± 0.00 | $0.56 | Rejected | no score |
| 65=GPT 5.4 (xhigh)OpenAI | 0.00%± 0.00 | $3.60 | Passes | 0 |
| 65=GPT 5.4 MiniOpenAI | 0.00%± 0.00 | $1.00 | Passes | 25 |
| 65=GPT 5.4 NanoOpenAI | 0.00%± 0.00 | $0.18 | Passes | 29 |
| 65=Grok 4.20 (Reasoning)xAI (SpaceXAI) | 0.00%± 0.00 | $0.99 | Rejected | no score |
| 65=Kimi K2.5Moonshot AI | 0.00%± 0.00 | $0.35 | Passes | 37 |
| 65=Laguna M.1Poolside | 0.00%± 0.00 | $0.53 | Rejected | no score |
| 65=Laguna XS.2Poolside | 0.00%± 0.00 | $0.28 | Rejected | no score |
| 65=Mercury 2.5Inception Labs | 0.00%± 0.00 | $0.19 | Not yet assessed | no score |
| 65=MiniMax-M2.7MiniMax | 0.00%± 0.00 | $0.49 | Rejected | no score |
| 65=Nemotron 3.5 LightningNVIDIA | 0.00%± 0.00 | $0.03 | Rejected | no score |
| 65=Qwen 3.7 PlusAlibaba | 0.00%± 0.00 | $0.23 | Not yet assessed | no score |
Source: Vals AI (Harvey LAB (Vals implementation)) · 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 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
- Legal Research Bench
- LegalBench (Vals leaderboard)
- CaseLaw v2
- TaxEval v2
- Tax Agent Bench
- CorpFin v2
- Immigration Bench
- Access to Justice
- Public Benefits Bench
- Vals Legal Index
- Vals Index
- GPQA Diamond (Vals-run)
- MMLU Pro (Vals-run)
- Artificial Analysis Intelligence Index
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.