Legal Research Bench
Professional Work · 2026-06-23
Vals AI Legal Research Bench evaluates legal research performance on private tasks spanning multiple practice areas.
Top models (higher is better)
| Model | Score |
|---|---|
| Claude Opus 5 | 55.3 |
| Claude Fable 5 | 49.5 |
| GPT-5.6 Sol | 48.1 |
| Kimi K3 | 44.2 |
| Opus 4.8 | 43.8 |
| Sonnet 5 | 41.8 |
| GPT-5.6 Terra | 40.9 |
| GPT-5.5 | 40.4 |
| Sonnet 4.6 | 38.5 |
| Grok 4.5 | 38.0 |
| Muse Spark 1.1 | 38.0 |
| GPT-5.6 Luna | 36.5 |
| GLM-5.2 | 31.3 |
| Gemini 3.5 Flash | 30.8 |
| MiniMax M3 | 29.8 |
| Inkling | 28.4 |
| GLM-5.1 | 27.9 |
| Inkling-Small | 25.5 |
| Qwen3.7-Max | 25.5 |
| Gemini 3.6 Flash | 25.0 |
| DeepSeek-V4-Pro | 23.1 |
| Gemini 3.1 Pro Preview | 20.7 |
| Kimi K2.6 | 15.9 |
| Grok 4.3 | 15.4 |
| Gemini 3.5 Flash-Lite | 13.9 |
| GPT-5.4 Mini | 12.5 |
| Laguna M.1 | 2.4 |
| Laguna XS.2 | 1.0 |