Benchmarks / Professional Domains
LegalBench
Legal reasoning task suite run by Vals AI.
maintained by Vals AI · www.vals.ai · 131 observations on record
Versions
1 version
| Version | Tasks |
|---|---|
| LegalBench current legal-bench | — |
Current leaderboard
LegalBench · Accuracy · top 25 of 131 systems
| # | System | Accuracy % |
|---|---|---|
| 1 | Claude Fable 5max effort Anthropic | 88.6% |
| 2 | Gemini 3.1 Pro Previewhigh effort Google DeepMind | 87.4% |
| 3 | Gemini 3 Pro Previewhigh effort Google DeepMind | 87.0% |
| 4 | GPT 5.6 Solmax effort OpenAI | 87.0% |
| 5 | Gemini 3 Flash Previewhigh effort Google DeepMind | 86.9% |
| 6 | GPT 5.5 (2026-04-22)xhigh effort OpenAI | 86.5% |
| 7 | GPT 5.4 (2026-03-05)xhigh effort OpenAI | 86.0% |
| 8 | GPT 5 (2025-08-07)high effort OpenAI | 86.0% |
| 9 | Grok 4.5high effort xAI | 86.0% |
| 10 | GPT 5.1 (2025-11-13)high effort OpenAI | 85.7% |
| 11 | Minimax M3 MiniMax | 85.4% |
| 12 | Claude Opus 4.6max effort Anthropic | 85.3% |
| 13 | Claude Opus 4.7max effort Anthropic | 85.3% |
| 14 | GPT 5.6 Terraxhigh effort OpenAI | 85.1% |
| 15 | GPT 5.6 Terramax effort OpenAI | 85.1% |
| 16 | Qwen3 5 Plusthinking Alibaba | 85.1% |
| 17 | Muse Spark 1.1xhigh effort Meta AI | 85.0% |
| 18 | Qwen3 7 Max Alibaba (Qwen) | 84.9% |
| 19 | Kimi K2 6 Moonshot AI | 84.7% |
| 20 | Claude Opus 4.5 20251101high effort Anthropic | 84.6% |
| 21 | Grok 4.3 xAI | 84.5% |
| 22 | GLM 5.1 Z.ai | 84.4% |
| 23 | GLM 5.1 Z.ai | 84.4% |
| 24 | Gemini 2.5 Pro Exp 03.25 Google DeepMind | 84.3% |
| 25 | Qwen3 5 Flash Alibaba | 84.3% |
One row per evaluated system — reasoning-effort variants rank separately.