Inference
EN
Vals AI

Legal Research Bench

Evaluating agents on legal research tasks across diverse areas of US law

ModelsModels with a result.90
Top resultBest result on this benchmark.55.8%Gemini 4 Carbon
Top-3 spreadPoints from first to third.0.5 pts
UpdatedDate the source changed the results.9 Nov 2026

Result and cost

89 of 90 results

Results

90 results
#ModelResultThe score from the source.CostUS dollars to run one task.TimeTime to run one task.
01Gemini 4 Carbon55.8% ±3.5$8.7452 min
02Claude Fable 5.155.3% ±3.5$23.0657 min
02Claude Opus 555.3% ±3.5$6.7631 min
02Muse Spark 1.3 Max55.3% ±3.5$0.5895 min
05Gemini 4 Argon54.8% ±3.5$6.4531 min
06Muse Wonder Full53.4% ±3.5$3.6028 min
07Claude Opus 5.550.5% ±3.5$23.611.9 h
07Muse Wonder50.5% ±3.5$3.9815 min
09Claude Fable 549.5% ±3.5$9.7922 min
10GLM-5.349.0% ±3.5$2.2452 min
10Muse Ember49.0% ±3.5$0.0573 min
12Claude Sonnet 5.548.1% ±3.5$13.661.6 h
12GPT-5.6 Sol48.1% ±3.5$21.6177 min
12Grok 4.648.1% ±3.5$1.5325 min
15Qwen3.8 Max47.6% ±3.5$2.4961 min
16MiMo-V2.6-Pro47.1% ±3.5$0.18130 min
16Grok 4.747.1% ±3.5$4.9221 min
18Inkling 2 Small46.6% ±3.5$0.28912 min
19Claude Fable 5.246.2% ±3.5$3.967 min
19Kimi K346.2% ±3.5$3.4713 min
19Qwen3.8 Max Internal46.2% ±3.5$5.312.1 h
22Hy4 preview45.2% ±3.5$0.84664 min
22GLM-5.3-Flash45.2% ±3.5$0.2451 min
24Ember-144.7% ±3.5$3.6739 min
25Claude Opus 4.843.8% ±3.4$2.8212 min
25Claude Haiku 5.543.8% ±3.4$1.6583 min
25Muse Spark 1.243.8% ±3.4$0.5085 min
28Claude Sonnet 541.8% ±3.4$2.7226 min
28GLM Legal41.8% ±3.4$0.04940 min
30Hy441.3% ±3.4$2.221.5 h
30GPT-5.6 Terra41.3% ±3.4$7.9170 min
30DeepSeek V4.1 Flash 101741.3% ±3.4$0.24812 min
33Muse Spark 1.340.9% ±3.4$0.659 min
33DeepSeek V4 Pro 081340.9% ±3.4$1.0942 min
35GPT-5.540.4% ±3.4$7.4038 min
35Step 5 Preview40.4% ±3.4$0.83936 min
37GPT-6 Astra39.4% ±3.4$10.5825 min
38Gemini 3.8 Flash38.9% ±3.4$1.635 min
39GPT-6.1 Sol38.5% ±3.4$3.3834 min
39Muse Spark Open38.5% ±3.4$1.5629 min
39Claude Opus 4.738.5% ±3.4$3.069 min
39Claude Sonnet 4.638.5% ±3.4$2.3016 min
43Muse Spark 1.138.0% ±3.4$0.3775 min
43MiMo-V2.6-Flash38.0% ±3.4$0.0715 min
43Grok 4.538.0% ±3.4$1.0916 min
46GPT-5.6 Luna36.5% ±3.3$0.84540 min
47Qwen3.8-27B36.1% ±3.3$2.1627 min
48Mistral Large 4 OSS35.6% ±3.3$1.0928 min
49Gemini 3.7 Flash34.6% ±3.3$0.765 min
50MiniMax M3.1 Flash34.1% ±3.3$0.44549 min
51Mistral Large 431.7% ±3.2$2.721.5 h
52GLM-5.231.3% ±3.2$0.88917 min
53Gemini 3.5 Flash30.8% ±3.2$1.315 min
54GPT-6 Luna30.3% ±3.2$0.4442 min
54DeepSeek V4 Flash 073130.3% ±3.2$0.27521 min
56MiniMax M329.8% ±3.2$0.34314 min
57GPT-6 Sol28.8% ±3.1$4.9024 min
57Beam 1.028.8% ±3.1—2 min
57Mercury 328.8% ±3.1$0.0747 min
60Inkling28.4% ±3.1$0.49414 min
61GLM-5.127.9% ±3.1$0.3865 min
62GPT-6.1 Terra26.0% ±3.0$3.2270 min
63Qwen3.7 Max25.5% ±3.0$1.4013 min
63Inkling-Small25.5% ±3.0$0.14213 min
65Gemini 3.6 Flash25.0% ±3.0$0.4673 min
66DeepSeek V4 Pro 042323.1% ±2.9$0.68217 min
67Gemini 3.1 Pro Preview20.7% ±2.8$1.104 min
68Gemini 3 Flash Preview18.3% ±2.7$0.1353 min
69Qwen3.7 Plus16.3% ±2.6$0.29614 min
70Kimi K2.615.9% ±2.5$1.5231 min
70MiMo-V2.5-Pro15.9% ±2.5$0.0837 min
70Kimi K2.5 Thinking15.9% ±2.5$0.2322 min
73Grok 4.315.4% ±2.5$0.20385 s
73Ling 3.0 Flash Af Rc315.4% ±2.5$0.0289 min
75Qwen3.6 Plus14.9% ±2.5$0.69811 min
76Grok 4.20 (Reasoning)13.9% ±2.4$0.2614 min
76Gemini 3.5 Flash-Lite13.9% ±2.4$0.12559 s
78GPT-5.4 mini12.5% ±2.3$1.8638 min
79MiniMax M2.710.6% ±2.1$0.2197 min
79Claude Haiku 4.5 Thinking10.6% ±2.1$0.2694 min
81Nemotron 3.5 Super VL 120B A12B10.1% ±2.1$0.06483 s
82MiMo-V2.59.1% ±2.0$0.0337 min
82Mistral Medium 3.59.1% ±2.0$1.3110 min
84GPT-5.4 nano6.3% ±1.7$0.1419 min
85Mercury 2.54.3% ±1.4$0.05848 s
86Gemini 3.1 Flash Lite Preview3.4% ±1.3$0.01420 s
87Nemotron Lightning 3p5 30B A3B2.9% ±1.2$0.00445 min
88Laguna M.12.4% ±1.1$0.228 min
89Laguna XS.21.0% ±0.7$0.1086 min
90Ling 3.0 Flash0.0%$0.0364 min