Legal Research Bench (Vals AI) leaderboard
- Claude Fable 5.1 leads
- 55.3%
- models measured
- 74
- at the top
- Tied
- median
- 37.3%
Leader on Legal Research Bench (Vals AI) · read October 22
On October 22, 2026, 74 of our catalog's models had a Legal Research Bench (Vals AI) result, and Anthropic Claude Fable 5.1 led with 55.3%.
US legal research by an agent, graded on its citations. The number is the share of tasks where every check passed.
Role in the index
Legal Research Bench (Vals AI) is shown as published and does not feed any of our indexes. Read the methodology
| # | Model | Result | Cost per test | Duration, as printed | Evidence | Coding index | Output $/Mtok |
|---|---|---|---|---|---|---|---|
| 1 | Claude Fable 5.1Anthropic | 55.3% | $23.06 | 57m 29s | 84.2 | $50.00 | |
| 2 | Claude Opus 5Anthropic | 55.3% | $6.76 | 31m 01s | 83.0 | $25.00 | |
| 3 | Gemini 4 ArgonGoogle | 54.8% | $6.45 | 31m 21s | 81.0 | $10.00 | |
| 4 | Claude Opus 5.5Anthropic | 50.5% | $25.16 | 1h 52m | 88.8 | $20.00 | |
| 5 | Muse WonderMeta | 50.5% | $3.98 | 15m 27s | 71.9 | $20.00 | |
| 6 | Claude Fable 5Anthropic | 49.5% | $9.79 | 22m 18s | 79.3 | $50.00 | |
| 7 | GLM 5.3Z.ai | 49.0% | $2.24 | 51m 50s | 70.6 | $4.40 | |
| 8 | Claude Sonnet 5.5Anthropic | 48.1% | $14.83 | 1h 33m | 87.3 | $10.00 | |
| 9 | GPT-5.6 SolOpenAI | 48.1% | $21.61 | 1h 17m | 73.5 | $20.00 | |
| 10 | Grok 4.6xAI | 48.1% | $1.53 | 25m 04s | 65.7 | $6.00 | |
| 11 | Qwen3.8 Max (0902)Qwen | 47.6% | $2.49 | 1h 01m | 62.4 | $6.00 | |
| 12 | Grok 4.7xAI | 47.1% | $4.92 | 21m 14s | 66.7 | $6.00 | |
| 13 | MiMo-V2.6-ProXiaomi | 47.1% | $0.18 | 30m 16s | 65.4 | $0.87 | |
| 14 | Kimi K3Moonshot AI | 46.1% | $3.47 | 12m 40s | 62.3 | $15.00 | |
| 15 | DeepSeek V4.1 Flash 1017DeepSeek | 45.2% | $0.70 | 35m 07s | 70.3 | $0.60 | |
| 16 | Hy4 previewTencent | 45.2% | $0.85 | 1h 04m | 44.3 | $2.501 | |
| 17 | GLM 5.3 FlashZ.ai | 45.2% | $0.12 | 51m 07s | 57.7 | $0.50 | |
| 18 | Ember-1Fireworks | 44.7% | $3.67 | 38m 34s | 52.6 | $15.00 | |
| 19 | Claude Opus 4.8Anthropic | 43.8% | $2.82 | 11m 54s | 65.0 | $25.00 | |
| 20 | Muse Spark 1.2Meta | 43.8% | $0.51 | 5m 07s | 55.7 | $4.25 | |
| 21 | Claude Haiku 5.5Anthropic | 43.3% | $1.63 | 1h 22m | 70.7 | $0.50 | |
| 22 | Claude Sonnet 5Anthropic | 41.8% | $2.72 | 25m 46s | 57.7 | $10.00 | |
| 23 | DeepSeek V4.1 FlashDeepSeek | 41.4% | $0.25 | 11m 42s | 62.5 | $0.84 | |
| 24 | GPT-5.6 TerraOpenAI | 41.4% | $7.91 | 1h 10m | 63.9 | $12.00 | |
| 25 | DeepSeek V4 Pro 0813DeepSeek | 40.9% | $1.09 | 42m 07s | 56.1 | $1.98 | |
| 26 | Muse Spark 1.3Meta | 40.9% | $0.65 | 8m 51s | 57.4 | $4.25 | |
| 27 | GPT-5.5OpenAI | 40.4% | $7.40 | 37m 45s | 71.3 | $30.00 | |
| 28 | Step 5 PreviewStepFun | 40.4% | $0.84 | 36m 08s | 61.1 | $2.70 | |
| 29 | GPT-6 AstraOpenAI | 39.4% | $10.58 | 24m 51s | 85.1 | $50.00 | |
| 30 | Gemini 3.8 FlashGoogle | 38.9% | $1.63 | 5m 24s | 61.6 | $3.75 | |
| 31 | Claude Opus 4.7Anthropic | 38.5% | $3.06 | 8m 32s | 67.5 | $25.00 | |
| 32 | Claude Sonnet 4.6Anthropic | 38.5% | $2.30 | 16m 30s | 57.3 | $15.00 | |
| 33 | Muse Spark OpenMeta | 38.5% | $1.56 | 29m 23s | 47.4 | $4.25 | |
| 34 | GPT-6.1 SolOpenAI | 38.5% | $3.38 | 34m 20s | 82.4 | $10.00 | |
| 35 | Muse Spark 1.1Meta | 38.0% | $0.38 | 5m 13s | 70.3 | $4.25 | |
| 36 | Grok 4.5xAI | 38.0% | $1.09 | 15m 54s | 58.6 | $6.00 | |
| 37 | MiMo-V2.6-FlashXiaomi | 38.0% | $0.07 | 15m 20s | 54.7 | $0.28 | |
| 38 | GPT-5.6 LunaOpenAI | 36.5% | $0.85 | 39m 34s | 58.0 | $1.20 | |
| 39 | Qwen3.8 27BQwen | 36.1% | $2.16 | 26m 41s | 74.2 | $2.55 | |
| 40 | Gemini 3.7 FlashGoogle | 34.6% | $0.76 | 4m 49s | 58.0 | $3.75 | |
| 41 | Mistral Large 4Mistral AI | 31.7% | $5.33 | 1h 33m | 56.4 | $4.18 | |
| 42 | GLM 5.2Z.ai | 31.3% | $0.89 | 17m 03s | 61.1 | $4.40 | |
| 43 | Gemini 3.5 FlashGoogle | 30.8% | $1.31 | 4m 38s | 39.8 | $9.00 | |
| 44 | DeepSeek V4 Flash 0731DeepSeek | 30.3% | $0.28 | 21m 07s | 51.8 | $0.396 | |
| 45 | GPT-6 LunaOpenAI | 30.3% | $0.44 | 42m 28s | 57.3 | $0.50 | |
| 46 | MiniMax M3MiniMax | 29.8% | $0.34 | 13m 34s | 39.2 | $1.20 | |
| 47 | GPT-6 SolOpenAI | 28.9% | $4.90 | 24m 09s | 76.0 | $10.00 | |
| 48 | T InklingThinking Machines | 28.4% | $0.49 | 14m 29s | 31.5 | $4.05 | |
| 49 | GLM 5.1Z.ai | 27.9% | $0.39 | 5m 26s | 49.2 | $4.40 | |
| 50 | Qwen3.7 MaxQwen | 25.5% | $1.40 | 13m 29s | 54.1 | $4.425 | |
| 51 | T Inkling SmallThinking Machines | 25.5% | $0.14 | 12m 55s | 27.3 | $1.20 | |
| 52 | Gemini 3.6 FlashGoogle | 25.0% | $0.47 | 2m 48s | 62.5 | $3.75 | |
| 53 | DeepSeek V4 Pro 0423DeepSeek | 23.1% | $0.68 | 17m 05s | 38.7 | $2.976 | |
| 54 | Gemini 3.1 Pro PreviewGoogle | 20.7% | $1.10 | 4m 02s | 32.3 | $12.00 | |
| 55 | Gemini 3 Flash PreviewGoogle | 18.3% | $0.14 | 3m 08s | 43.7 | $3.00 | |
| 56 | Qwen3.7 PlusQwen | 16.4% | $0.30 | 13m 52s | 33.2 | $1.28 | |
| 57 | Kimi K2.5Moonshot AI | 15.9% | $0.23 | 22m 13s | 38.6 | $2.85 | |
| 58 | Kimi K2.6Moonshot AI | 15.9% | $1.52 | 30m 53s | 52.1 | $4.00 | |
| 59 | MiMo-V2.5-ProXiaomi | 15.9% | $0.08 | 6m 56s | 31.6 | $0.9152 | |
| 60 | Nemotron 3 UltraNVIDIA | 15.4% | Not published | 10m 11s | 30.0 | $2.40 | |
| 61 | Grok 4.3xAI | 15.4% | $0.20 | 1m 25s | 37.4 | $2.50 | |
| 62 | Qwen3.6 PlusQwen | 14.9% | $0.70 | 10m 59s | 44.1 | $1.95 | |
| 63 | Gemini 3.5 Flash LiteGoogle | 13.9% | $0.12 | 59 s | 49.0 | $2.50 | |
| 64 | Grok 4.20xAI | 13.9% | $0.26 | 4m 04s | 33.3 | $2.50 | |
| 65 | GPT-5.4 MiniOpenAI | 12.5% | $1.86 | 37m 36s | 35.8 | $4.50 | |
| 66 | Claude Haiku 4.5Anthropic | 10.6% | $0.27 | 4m 03s | 35.6 | $5.00 | |
| 67 | MiniMax M2.7MiniMax | 10.6% | $0.22 | 7m 01s | 36.2 | $1.20 | |
| 68 | Mistral Medium 3.5Mistral AI | 9.1% | $1.31 | 10m 13s | 25.5 | $7.50 | |
| 69 | MiMo-V2.5Xiaomi | 9.1% | $0.03 | 6m 31s | 50.0 | $0.336 | |
| 70 | GPT-5.4 NanoOpenAI | 6.3% | $0.14 | 8m 47s | 44.8 | $1.25 | |
| 71 | Mercury 2.5Inception | 4.3% | $0.06 | 48 s | 1.6 | $0.15 | |
| 72 | Gemini 3.1 Flash Lite PreviewGoogle | 3.4% | $0.01 | 20 s | 22.7 | $1.50 | |
| 73 | Nemotron 3.5 LightningNVIDIA | 2.9% | $0.00 | 4m 45s | 22.6 | $0.147 | |
| 74 | Ling 3.0 FlashinclusionAI | 0.0% | $0.04 | 3m 32s | 29.6 | $0.1215 |
Only models in our catalog are listed, and each appears once with its best published run. 137 rows were read from the source on October 22, 2026, and rows that match no catalog model are left out. Cost per test is what Vals AI reports one test cost on the run behind the result, in US dollars. Every other column is a figure the board publishes for the same run.