Benchmarks · Vals AI
GPQA Diamond (Vals-run)
General reasoning benchmark: graduate-level science multiple-choice questions (run by Vals; "Archived Benchmark", saturated, no new models; last update 1 September 2026). Not legal.
- Models measured
- 109
- Top
- 95.45 %Gemini 3.1 Pro Preview (02/26)within the standard error of ±1.05
- Version
- 1
- Measured
- 1 September 2026
01
Ranking
At the top: Top 15
Vals AI Legal Research Bench, all-pass [1]. Hatched bars are rejected by the compliance gate.
- 1Gemini 3.1 Pro Preview (02/26)95.45%
- 2GPT-5.6 Sol95.20%
- 3Grok 4.694.70%
- 4Gemini 3.8 Flash94.44%
- 5Gemini 3.7 Flash93.94%
- 6Qwen 3.8 Max93.69%
- 7=Claude Fable 5.193.43%
- 7=Claude Opus 593.43%
- 7=Gemini 3.6 Flash93.43%
- 10=Claude Fable 593.18%
- 10=GPT 5.593.18%
- 12=Grok 4.592.93%
- 12=Kimi K392.93%
- 14=Gemini 3.5 Flash92.68%
- 14=MiniMax-M392.68%
Source: Vals AI (GPQA Diamond (Vals-run)) · 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 |
|---|---|---|---|---|
| 1Gemini 3.1 Pro Preview (02/26)Google | 95.45%± 1.05 | - | Rejected | no score |
| 2GPT-5.6 SolOpenAI | 95.20%± 1.07 | - | Passes | 57 |
| 3Grok 4.6xAI (SpaceXAI) | 94.70%± 1.13 | - | Rejected | no score |
| 4Gemini 3.8 FlashGoogle | 94.44%± 1.48 | - | Passes | 62 |
| 5Gemini 3.7 FlashGoogle | 93.94%± 1.49 | - | Passes | 58 |
| 6Qwen 3.8 MaxAlibaba | 93.69%± 1.22 | - | Not yet assessed | no score |
| 7=Claude Fable 5.1Anthropic | 93.43%± 1.86 | - | Rejected | no score |
| 7=Claude Opus 5Anthropic | 93.43%± 1.24 | - | Passes | 72 |
| 7=Gemini 3.6 FlashGoogle | 93.43%± 1.33 | - | Passes | 46 |
| 10=Claude Fable 5Anthropic | 93.18%± 1.94 | - | Rejected | no score |
| 10=GPT 5.5OpenAI | 93.18%± 1.29 | - | Passes | 51 |
| 12=Grok 4.5xAI (SpaceXAI) | 92.93%± 1.29 | - | Rejected | no score |
| 12=Kimi K3Moonshot AI | 92.93%± 1.31 | - | Rejected | no score |
| 14=Gemini 3.5 FlashGoogle | 92.68%± 1.46 | - | Passes | 47 |
| 14=MiniMax-M3MiniMax | 92.68%± 1.44 | - | Rejected | no score |
| 16=Claude Opus 4.8Anthropic | 92.42%± 1.98 | - | Passes | 62 |
| 16=DeepSeek V4 Pro (0813)DeepSeek | 92.42%± 2.02 | - | Rejected | no score |
| 18=Gemini 3 Pro (11/25)Google | 91.67%± 1.39 | - | Rejected | no score |
| 18=GPT 5.2OpenAI | 91.67%± 1.84 | - | Passes | no score |
| 18=GPT 5.4 (xhigh)OpenAI | 91.67%± 1.91 | - | Passes | 0 |
| 18=GPT-5.6 LunaOpenAI | 91.67%± 1.74 | - | Passes | 56 |
| 22Grok 4.3xAI (SpaceXAI) | 91.41%± 1.41 | - | Rejected | no score |
| 23Muse Spark 1.1Meta | 91.16%± 2.10 | - | Rejected | no score |
| 24GPT-5.6 TerraOpenAI | 90.91%± 1.91 | - | Passes | 53 |
| 25=Claude Opus 4.7Anthropic | 90.15%± 1.54 | - | Passes | 54 |
| 25=Qwen 3.7 MaxAlibaba | 90.15%± 1.50 | - | Not yet assessed | no score |
| 27DeepSeek V4 Flash 0731DeepSeek | 89.90%± 1.69 | - | Rejected | no score |
| 28=Claude Opus 4.6 (Thinking)Anthropic | 89.65%± 1.19 | - | Passes | no score |
| 28=Muse SparkMeta | 89.65%± 1.55 | - | Rejected | no score |
| 30DeepSeek V4 Pro (2026-04-23)DeepSeek | 89.39%± 1.65 | - | Rejected | no score |
| 31Kimi K2.6Moonshot AI | 89.14%± 2.00 | - | Rejected | no score |
| 32=Claude Sonnet 5Anthropic | 88.89%± 2.22 | - | Passes | 59 |
| 32=Qwen 3.8 27BAlibaba | 88.89%± 1.58 | - | Rejected | no score |
| 34Grok 4.20 (Reasoning)xAI (SpaceXAI) | 88.64%± 1.59 | - | Rejected | no score |
| 35=GLM 5.3Z.ai | 88.13%± 1.85 | - | Rejected | no score |
| 35=Grok 4xAI (SpaceXAI) | 88.13%± 1.63 | - | Rejected | no score |
| 37Gemini 3 Flash (12/25)Google | 87.88%± 1.64 | - | Rejected | no score |
| 38=Qwen 3.5 PlusAlibaba | 87.37%± 1.67 | - | Not yet assessed | no score |
| 38=Qwen 3.6 PlusAlibaba | 87.37%± 1.67 | - | Not yet assessed | no score |
| 40InklingThinking Machines | 87.12%± 1.78 | - | Rejected | no score |
| 41=GPT 5.1OpenAI | 86.62%± 1.71 | - | Passes | no score |
| 41=MiniMax-M2.7MiniMax | 86.62%± 1.95 | - | Rejected | no score |
| 43GLM 5.3 FlashZ.ai | 86.36%± 2.13 | - | Rejected | no score |
| 44Nemotron 3 UltraNVIDIA | 86.11%± 1.74 | - | Rejected | no score |
| 45Claude Opus 4.5 (Thinking)Anthropic | 85.86%± 2.48 | - | Passes | no score |
| 46=Claude Sonnet 4.6Anthropic | 85.61%± 2.41 | - | Passes | 55 |
| 46=GLM 5.2Z.ai | 85.61%± 1.77 | - | Rejected | no score |
| 46=GPT 5OpenAI | 85.61%± 1.76 | - | Passes | no score |
| 49Grok 4 Fast (Reasoning)xAI (SpaceXAI) | 85.35%± 1.79 | - | Rejected | no score |
| 50=Ling 3.0 FlashAnt Group | 84.85%± 2.20 | - | Rejected | no score |
| 50=Qwen 3 Max ThinkingAlibaba | 84.85%± 1.80 | - | Not yet assessed | no score |
| 52GLM 5.1Z.ai | 84.52%± 1.82 | - | Rejected | no score |
| 53Grok 4.1 Fast (Reasoning)xAI (SpaceXAI) | 84.34%± 1.85 | - | Rejected | no score |
| 54=Kimi K2.5Moonshot AI | 84.09%± 2.13 | - | Passes | 37 |
| 54=o3OpenAI | 84.09%± 1.86 | - | Passes | no score |
| 56Gemini 3.5 Flash LiteGoogle | 83.84%± 1.85 | - | Passes | 38 |
| 57Inkling SmallThinking Machines | 83.59%± 2.04 | - | Rejected | no score |
| 58GLM 5Z.ai | 83.33%± 1.87 | - | Passes | no score |
| 59GPT 5.4 MiniOpenAI | 83.08%± 2.46 | - | Passes | 25 |
| 60Qwen 3.5 FlashAlibaba | 82.83%± 1.90 | - | Not yet assessed | no score |
| 61MiMo V2.5 ProXiaomi | 82.58%± 2.47 | - | Rejected | no score |
| 62MiniMax-M2.5MiniMax | 82.07%± 2.26 | - | Passes | no score |
| 63Claude Sonnet 4.5 (Thinking)Anthropic | 81.63%± 2.25 | - | Passes | no score |
| 64=Gemini 2.5 Flash Preview (9/25) (Nonthinking)Google | 81.57%± 2.00 | - | Rejected | no score |
| 64=MiMo V2.5Xiaomi | 81.57%± 2.05 | - | Rejected | no score |
| 66Gemini 3.1 Flash Lite PreviewGoogle | 81.06%± 1.97 | - | Not yet assessed | no score |
| 67=DeepSeek V3.2 (Thinking)DeepSeek | 80.30%± 2.00 | - | Passes | no score |
| 67=GPT 5 MiniOpenAI | 80.30%± 2.00 | - | Passes | no score |
| 69GLM 4.7Z.ai | 80.05%± 2.01 | - | Passes | no score |
| 70=Claude Opus 4.5 (Nonthinking)Anthropic | 79.55%± 2.46 | - | Passes | no score |
| 70=Qwen 3 MaxAlibaba | 79.55%± 2.03 | - | Not yet assessed | no score |
| 72=GPT OSS 120BOpenAI | 78.54%± 2.25 | - | Passes | no score |
| 72=Kimi K2 ThinkingMoonshot AI | 78.54%± 2.18 | - | Rejected | no score |
| 72=MiniMax-M2.1MiniMax | 78.54%± 2.13 | - | Passes | no score |
| 75Qwen 3 Max PreviewAlibaba | 77.78%± 2.09 | - | Not yet assessed | no score |
| 76GPT 5.4 NanoOpenAI | 77.53%± 2.42 | - | Passes | 29 |
| 77Gemini 2.5 Flash Preview (9/25) (Thinking)Google | 76.52%± 2.15 | - | Rejected | no score |
| 78=Claude Opus 4.1 (Thinking)Anthropic | 76.26%± 2.16 | - | Rejected | no score |
| 78=DeepSeek V3.2 (Nonthinking)DeepSeek | 76.26%± 2.16 | - | Passes | no score |
| 80Claude Sonnet 4 (Thinking)Anthropic | 75.00%± 2.19 | - | Rejected | no score |
| 81=GLM 4.6Z.ai | 74.50%± 2.20 | - | Rejected | no score |
| 81=o4 MiniOpenAI | 74.50%± 2.19 | - | Passes | no score |
| 83=Claude Haiku 4.5 (Thinking)Anthropic | 72.22%± 2.47 | - | Passes | 32 |
| 83=GLM 4.5Z.ai | 72.22%± 2.26 | - | Rejected | no score |
| 85Claude Opus 4 (Nonthinking)Anthropic | 71.72%± 2.26 | - | Rejected | no score |
| 86Kimi K2 InstructMoonshot AI | 71.46%± 2.27 | - | Rejected | no score |
| 87Gemini 2.5 Flash Lite (9/25) (Thinking)Google | 70.20%± 2.30 | - | Rejected | no score |
| 88Claude Opus 4.1 (Nonthinking)Anthropic | 69.95%± 2.30 | - | Rejected | no score |
| 89=Claude Sonnet 4 (Nonthinking)Anthropic | 69.44%± 2.31 | - | Rejected | no score |
| 89=Llama 4 MaverickMeta | 69.44%± 2.35 | - | Rejected | no score |
| 91GPT OSS 20BOpenAI | 68.94%± 2.46 | - | Passes | no score |
| 92Mistral Large 3Mistral AI | 68.43%± 2.37 | - | Passes | no score |
| 93GPT 4.1 MiniOpenAI | 67.93%± 2.35 | - | Passes | no score |
| 94GPT 4.1OpenAI | 65.40%± 2.40 | - | Passes | no score |
| 95Grok 4.1 Fast Non-ReasoningxAI (SpaceXAI) | 65.15%± 2.40 | - | Rejected | no score |
| 96Gemini 2.5 Flash Lite (9/25) (Nonthinking)Google | 64.14%± 2.41 | - | Rejected | no score |
| 97GPT 5 NanoOpenAI | 63.38%± 2.46 | - | Passes | no score |
| 98Magistral Medium 1.2 (09/2025)Mistral AI | 62.37%± 2.44 | - | Not yet assessed | no score |
| 99Grok 4 Fast (Non-Reasoning)xAI (SpaceXAI) | 62.12%± 2.43 | - | Rejected | no score |
| 100MiMo V2 FlashXiaomi | 59.34%± 2.50 | - | Rejected | no score |
| 101Magistral Small 1.2 (09/2025)Mistral AI | 58.33%± 2.49 | - | Passes | no score |
| 102Laguna XS.2Poolside | 55.05%± 2.50 | - | Rejected | no score |
| 103GPT 4o (2024-11-20)OpenAI | 53.79%± 2.51 | - | Passes | no score |
| 104GPT 4.1 NanoOpenAI | 50.76%± 2.46 | - | Passes | no score |
| 105Command ACohere | 48.48%± 2.28 | - | Rejected | no score |
| 106Llama 4 ScoutMeta | 46.97%± 2.49 | - | Rejected | no score |
| 107Mistral Small 3.1 (03/2025)Mistral AI | 44.19%± 2.47 | - | Passes | no score |
| 108Mistral Medium 3.5Mistral AI | 34.85%± 2.39 | - | Passes | 23 |
| 109Laguna M.1Poolside | 27.02%± 1.92 | - | Rejected | no score |
Source: Vals AI (GPQA Diamond (Vals-run)) · 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 1 September 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
- Access to Justice
- Public Benefits Bench
- Vals Legal Index
- Vals Index
- 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.