Benchmarks · Vals AI
Access to Justice
Measures to what extent AI agents can guide an unrepresented client through a real legal problem.
- Models measured
- 23
- Top
- 76.37 %Qwen 3.8 Maxwithin the standard error of ±0.68
- Version
- 1
- Measured
- 10 November 2026
01
Ranking
At the top: Top 15
Vals AI Legal Research Bench, all-pass [1]. Hatched bars are rejected by the compliance gate.
- 1Qwen 3.8 Max76.37%
- 2Claude Sonnet 5.574.94%
- 3Wonder Full74.68%
- 4Claude Opus 5.573.77%
- 5GLM 5.373.70%
- 6Grok 4.773.69%
- 7MiniMax-M3.1-Flash73.31%
- 8DeepSeek V4.1 Flash 101773.11%
- 9Wonder72.96%
- 10DeepSeek V4.1 Flash72.86%
- 11Muse Spark 1.372.18%
- 12Muse Spark 1.3 (Max)72.15%
- 13Grok 4.671.34%
- 14GLM 5.3 Flash71.11%
- 15MiMo V2.6 Flash70.75%
Source: Vals AI (Access to Justice) · 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 |
|---|---|---|---|---|
| 1Qwen 3.8 MaxAlibaba | 76.37%± 0.68 | $0.61 | Not yet assessed | no score |
| 2Claude Sonnet 5.5Anthropic | 74.94%± 0.70 | $2.37 | Passes | 62 |
| 3Wonder FullMeta | 74.68%± 0.70 | $1.24 | Rejected | no score |
| 4Claude Opus 5.5Anthropic | 73.77%± 0.71 | $3.38 | Passes | 63 |
| 5GLM 5.3Z.ai | 73.70%± 0.71 | $0.46 | Rejected | no score |
| 6Grok 4.7xAI (SpaceXAI) | 73.69%± 0.71 | $0.82 | Rejected | no score |
| 7MiniMax-M3.1-FlashMiniMax | 73.31%± 0.71 | $0.06 | Rejected | no score |
| 8DeepSeek V4.1 Flash 1017DeepSeek | 73.11%± 0.71 | $0.09 | Rejected | no score |
| 9WonderMeta | 72.96%± 0.72 | $1.10 | Rejected | no score |
| 10DeepSeek V4.1 FlashDeepSeek | 72.86%± 0.72 | $0.11 | Rejected | no score |
| 11Muse Spark 1.3Meta | 72.18%± 0.72 | $0.18 | Rejected | no score |
| 12Muse Spark 1.3 (Max)Meta | 72.15%± 0.72 | $0.22 | Rejected | no score |
| 13Grok 4.6xAI (SpaceXAI) | 71.34%± 0.73 | $0.71 | Rejected | no score |
| 14GLM 5.3 FlashZ.ai | 71.11%± 0.73 | $0.02 | Rejected | no score |
| 15MiMo V2.6 FlashXiaomi | 70.75%± 0.73 | $0.02 | Rejected | no score |
| 16MiMo V2.6 ProXiaomi | 69.96%± 0.74 | $0.05 | Rejected | no score |
| 17MiniMax-M3MiniMax | 69.77%± 0.74 | $0.11 | Rejected | no score |
| 18Kimi K3.1 ProMoonshot AI | 69.47%± 0.74 | $0.96 | Rejected | no score |
| 19Gemini 4 ArgonGoogle | 68.48%± 0.75 | $0.83 | Passes | 79 |
| 20Kimi K3Moonshot AI | 68.17%± 0.75 | $0.45 | Rejected | no score |
| 21Gemini 3.8 FlashGoogle | 67.06%± 0.76 | $0.32 | Passes | 62 |
| 22GPT-6.1 SolOpenAI | 65.84%± 0.76 | $0.69 | Passes | 59 |
| 23GPT-5.6 LunaOpenAI | 65.56%± 0.77 | $0.09 | Passes | 56 |
Source: Vals AI (Access to Justice) · 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 10 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
- Legal Research Bench
- Harvey LAB (Vals implementation)
- LegalBench (Vals leaderboard)
- CaseLaw v2
- TaxEval v2
- Tax Agent Bench
- CorpFin v2
- Immigration Bench
- 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.