Benchmarks · Vals AI
CorpFin v2
Measures understanding of long credit agreements (question answering over contract text; private dataset). Vals marks the leaderboard "Archived Benchmark" (saturated; no new models; last update 12 August 2026).
- Models measured
- 108
- Top
- 73.19 %Claude Opus 5within the standard error of ±0.87
- Version
- 2
- Measured
- 12 August 2026
01
Ranking
At the top: Top 15
Vals AI Legal Research Bench, all-pass [1]. Hatched bars are rejected by the compliance gate.
- 1Claude Opus 573.19%
- 2Claude Fable 571.83%
- 3Kimi K371.56%
- 4Muse Spark 1.171.29%
- 5Muse Spark 1.270.94%
- 6Inkling Small69.62%
- 7Inkling68.57%
- 8Grok 4.368.53%
- 9GPT 5.568.42%
- 10Kimi K2.568.26%
- 11MiniMax-M368.10%
- 12Qwen 3 Max Thinking68.03%
- 13Grok 4.567.41%
- 14Claude Opus 4.6 (Thinking)67.02%
- 15Claude Sonnet 566.98%
Source: Vals AI (CorpFin v2) · 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 |
|---|---|---|---|---|
| 1Claude Opus 5Anthropic | 73.19%± 0.87 | - | Passes | 72 |
| 2Claude Fable 5Anthropic | 71.83%± 0.88 | - | Rejected | no score |
| 3Kimi K3Moonshot AI | 71.56%± 0.89 | - | Rejected | no score |
| 4Muse Spark 1.1Meta | 71.29%± 0.89 | - | Rejected | no score |
| 5Muse Spark 1.2Meta | 70.94%± 0.89 | - | Rejected | no score |
| 6Inkling SmallThinking Machines | 69.62%± 0.91 | - | Rejected | no score |
| 7InklingThinking Machines | 68.57%± 0.92 | - | Rejected | no score |
| 8Grok 4.3xAI (SpaceXAI) | 68.53%± 0.92 | - | Rejected | no score |
| 9GPT 5.5OpenAI | 68.42%± 0.92 | - | Passes | 51 |
| 10Kimi K2.5Moonshot AI | 68.26%± 0.92 | - | Passes | 37 |
| 11MiniMax-M3MiniMax | 68.10%± 0.92 | - | Rejected | no score |
| 12Qwen 3 Max ThinkingAlibaba | 68.03%± 0.92 | - | Not yet assessed | no score |
| 13Grok 4.5xAI (SpaceXAI) | 67.41%± 0.92 | - | Rejected | no score |
| 14Claude Opus 4.6 (Thinking)Anthropic | 67.02%± 0.93 | - | Passes | no score |
| 15Claude Sonnet 5Anthropic | 66.98%± 0.93 | - | Passes | 59 |
| 16Grok 4 Fast (Reasoning)xAI (SpaceXAI) | 66.90%± 0.93 | - | Rejected | no score |
| 17Kimi K2.6Moonshot AI | 66.74%± 0.93 | - | Rejected | no score |
| 18Claude Opus 4.8Anthropic | 66.71%± 0.93 | - | Passes | 62 |
| 19Qwen 3.6 Max PreviewAlibaba | 66.47%± 0.93 | - | Not yet assessed | no score |
| 20Gemini 3 Flash (12/25)Google | 66.43%± 0.93 | - | Rejected | no score |
| 21Grok 4.6xAI (SpaceXAI) | 66.16%± 0.93 | - | Rejected | no score |
| 22GLM 5.2Z.ai | 66.12%± 0.93 | - | Rejected | no score |
| 23Claude Opus 4.7Anthropic | 66.08%± 0.93 | - | Passes | 54 |
| 24Grok 4xAI (SpaceXAI) | 66.05%± 0.93 | - | Rejected | no score |
| 25Grok 4.1 Fast (Reasoning)xAI (SpaceXAI) | 65.97%± 0.93 | - | Rejected | no score |
| 26GPT 5.2OpenAI | 65.89%± 0.93 | - | Passes | no score |
| 27Qwen 3.8 MaxAlibaba | 65.85%± 0.94 | - | Not yet assessed | no score |
| 28Nemotron 3 UltraNVIDIA | 65.46%± 0.94 | - | Rejected | no score |
| 29DeepSeek V4 Pro (0813)DeepSeek | 65.42%± 0.94 | - | Rejected | no score |
| 30=Claude Sonnet 4.6Anthropic | 65.31%± 0.94 | - | Passes | 55 |
| 30=GPT-5.6 TerraOpenAI | 65.31%± 0.94 | - | Passes | 53 |
| 30=Qwen 3.5 PlusAlibaba | 65.31%± 0.94 | - | Not yet assessed | no score |
| 33GPT 5.4 (xhigh)OpenAI | 65.27%± 0.94 | - | Passes | 0 |
| 34Muse SparkMeta | 65.11%± 0.94 | - | Rejected | no score |
| 35Claude Opus 4.5 (Thinking)Anthropic | 65.07%± 0.94 | - | Passes | no score |
| 36Gemini 3.5 FlashGoogle | 64.69%± 0.94 | - | Passes | 47 |
| 37Gemini 3.1 Pro Preview (02/26)Google | 64.49%± 0.94 | - | Rejected | no score |
| 38GLM 5.1Z.ai | 64.45%± 0.94 | - | Rejected | no score |
| 39GPT-5.6 SolOpenAI | 64.38%± 0.94 | - | Passes | 57 |
| 40GPT-5.6 LunaOpenAI | 64.22%± 0.94 | - | Passes | 56 |
| 41GPT 5.1OpenAI | 63.83%± 0.95 | - | Passes | no score |
| 42Qwen 3.7 MaxAlibaba | 63.71%± 0.95 | - | Not yet assessed | no score |
| 43=Gemini 3 Pro (11/25)Google | 63.67%± 0.95 | - | Rejected | no score |
| 43=Grok 4.20 (Reasoning)xAI (SpaceXAI) | 63.67%± 0.95 | - | Rejected | no score |
| 45Qwen 3.5 FlashAlibaba | 63.56%± 0.95 | - | Not yet assessed | no score |
| 46Gemini 3.6 FlashGoogle | 63.33%± 0.95 | - | Passes | 46 |
| 47GPT 4.1OpenAI | 63.05%± 0.95 | - | Passes | no score |
| 48GLM 5Z.ai | 62.90%± 0.95 | - | Passes | no score |
| 49Qwen 3.6 27BAlibaba | 62.31%± 0.95 | - | Rejected | no score |
| 50Claude Sonnet 4.5 (Thinking)Anthropic | 61.97%± 0.96 | - | Passes | no score |
| 51=Ling 3.0 FlashAnt Group | 61.93%± 0.96 | - | Rejected | no score |
| 51=Qwen 3.6 PlusAlibaba | 61.93%± 0.96 | - | Not yet assessed | no score |
| 53DeepSeek V4 Flash 0731DeepSeek | 61.85%± 0.96 | - | Rejected | no score |
| 54MiMo V2.5 ProXiaomi | 61.42%± 0.96 | - | Rejected | no score |
| 55DeepSeek V4 Pro (2026-04-23)DeepSeek | 61.38%± 0.96 | - | Rejected | no score |
| 56Claude Opus 4.5 (Nonthinking)Anthropic | 61.30%± 0.96 | - | Passes | no score |
| 57Claude Sonnet 4 (Thinking)Anthropic | 61.23%± 0.96 | - | Rejected | no score |
| 58=GPT 5.4 NanoOpenAI | 61.19%± 0.96 | - | Passes | 29 |
| 58=MiniMax-M2.7MiniMax | 61.19%± 0.96 | - | Rejected | no score |
| 60GPT 5OpenAI | 61.07%± 0.96 | - | Passes | no score |
| 61Mistral Large 3Mistral AI | 61.03%± 0.96 | - | Passes | no score |
| 62GLM 4.5Z.ai | 60.96%± 0.96 | - | Rejected | no score |
| 63GPT 5.4 MiniOpenAI | 60.92%± 0.96 | - | Passes | 25 |
| 64=Claude Sonnet 4.5 (Nonthinking)Anthropic | 60.80%± 0.96 | - | Passes | no score |
| 64=Gemini 2.5 ProGoogle | 60.80%± 0.96 | - | Not yet assessed | no score |
| 66Gemini 3.5 Flash LiteGoogle | 60.68%± 0.96 | - | Passes | 38 |
| 67Claude Haiku 4.5 (Thinking)Anthropic | 60.61%± 0.96 | - | Passes | 32 |
| 68Kimi K2 ThinkingMoonshot AI | 60.57%± 0.96 | - | Rejected | no score |
| 69Claude Haiku 4.5 (Nonthinking)Anthropic | 60.30%± 0.96 | - | Passes | no score |
| 70GPT 5 MiniOpenAI | 60.18%± 0.96 | - | Passes | no score |
| 71MiMo V2.5Xiaomi | 59.91%± 0.97 | - | Rejected | no score |
| 72Gemini 2.5 Flash Preview (9/25) (Thinking)Google | 59.75%± 0.97 | - | Rejected | no score |
| 73o3OpenAI | 59.71%± 0.96 | - | Passes | no score |
| 74MiniMax-M2.5MiniMax | 59.60%± 0.97 | - | Passes | no score |
| 75Gemini 3.1 Flash Lite PreviewGoogle | 59.36%± 0.97 | - | Not yet assessed | no score |
| 76=Gemini 2.5 Flash Preview (9/25) (Nonthinking)Google | 58.97%± 0.97 | - | Rejected | no score |
| 76=o4 MiniOpenAI | 58.97%± 0.97 | - | Passes | no score |
| 78MiniMax-M2.1MiniMax | 58.90%± 0.97 | - | Passes | no score |
| 79Mistral Medium 3.5Mistral AI | 58.78%± 0.97 | - | Passes | 23 |
| 80Grok 4 Fast (Non-Reasoning)xAI (SpaceXAI) | 58.39%± 0.97 | - | Rejected | no score |
| 81GPT OSS 120BOpenAI | 58.24%± 0.97 | - | Passes | no score |
| 82Laguna M.1Poolside | 58.16%± 0.97 | - | Rejected | no score |
| 83GPT 4.1 MiniOpenAI | 57.93%± 0.97 | - | Passes | no score |
| 84Gemini 2.5 Flash Lite (9/25) (Thinking)Google | 57.58%± 0.97 | - | Rejected | no score |
| 85GLM 4.6Z.ai | 56.84%± 0.97 | - | Rejected | no score |
| 86Laguna XS.2Poolside | 56.33%± 0.98 | - | Rejected | no score |
| 87Gemini 2.5 Flash Lite (9/25) (Nonthinking)Google | 56.29%± 0.98 | - | Rejected | no score |
| 88Qwen 3 MaxAlibaba | 55.94%± 0.98 | - | Not yet assessed | no score |
| 89Claude Sonnet 4 (Nonthinking)Anthropic | 54.70%± 0.98 | - | Rejected | no score |
| 90Trinity Large ThinkingArcee AI | 54.66%± 0.98 | - | Rejected | no score |
| 91Nemotron 3.5 LightningNVIDIA | 54.23%± 0.98 | - | Rejected | no score |
| 92GPT OSS 20BOpenAI | 53.15%± 0.98 | - | Passes | no score |
| 93Qwen 3 Max PreviewAlibaba | 52.95%± 0.97 | - | Not yet assessed | no score |
| 94Grok 4.1 Fast Non-ReasoningxAI (SpaceXAI) | 52.49%± 0.97 | - | Rejected | no score |
| 95DeepSeek V3.1DeepSeek | 51.48%± 0.98 | - | Passes | no score |
| 96DeepSeek V3.2 (Thinking)DeepSeek | 50.97%± 0.98 | - | Passes | no score |
| 97Mistral Medium 3.1 (05/2025)Mistral AI | 50.74%± 0.97 | - | Passes | no score |
| 98Kimi K2 InstructMoonshot AI | 50.39%± 0.98 | - | Rejected | no score |
| 99Llama 4 MaverickMeta | 49.73%± 0.98 | - | Rejected | no score |
| 100DeepSeek V3.2 (Nonthinking)DeepSeek | 47.94%± 0.98 | - | Passes | no score |
| 101Magistral Medium 1.2 (09/2025)Mistral AI | 47.40%± 0.98 | - | Not yet assessed | no score |
| 102Llama 4 ScoutMeta | 46.78%± 0.98 | - | Rejected | no score |
| 103GLM 4.7Z.ai | 46.39%± 0.90 | - | Passes | no score |
| 104Command ACohere | 45.96%± 0.97 | - | Rejected | no score |
| 105GPT 4o (2024-11-20)OpenAI | 45.92%± 0.97 | - | Passes | no score |
| 106Mistral Small 3.1 (03/2025)Mistral AI | 44.17%± 0.97 | - | Passes | no score |
| 107Magistral Small 1.2 (09/2025)Mistral AI | 44.02%± 0.96 | - | Passes | no score |
| 108GPT 4.1 NanoOpenAI | 42.08%± 0.97 | - | Passes | no score |
Source: Vals AI (CorpFin v2) · 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 · Measured 12 August 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
- 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.