LegalBench - Rule Tasks
Professional Work · 2023-08-18
The Rule Tasks split of LegalBench. This child benchmark separates a source-reported subtask or subtrack from the parent aggregate so scores at different grains do not share one benchmark_slug.
Top models (higher is better)
| Model | Score |
|---|---|
| Gemini 4 Argon | 90.8 |
| Claude Fable 5 | 88.3 |
| Gemini 3.1 Pro Preview | 88.3 |
| Claude Fable 5.1 | 88.1 |
| Gemini 3.5 Flash | 88.0 |
| Muse Wonder Full | 87.1 |
| Gemini 3.8 Flash | 87.1 |
| GPT-5.6 Sol | 86.4 |
| Gemini 3.7 Flash | 85.9 |
| Gemini 3.6 Flash | 85.7 |
| GPT-5 | 85.7 |
| GPT-5.5 | 85.5 |
| Gemini 3 Flash Preview | 85.3 |
| Gemini 3 Pro Preview | 85.2 |
| Opus 4.7 | 85.0 |
| Claude Opus 5 | 84.8 |
| Muse Spark 1.2 | 84.7 |
| Grok 4.6 | 84.5 |
| Grok 4.5 | 84.2 |
| Kimi K3 | 84.0 |
| GPT-5.4 | 84.0 |
| GPT-5.1 | 83.6 |
| Muse Spark 1.1 | 83.6 |
| Opus 4.6 | 82.4 |
| Gemini 3.1 Flash-Lite Preview | 82.2 |
| Claude 3.7 Sonnet | 82.1 |
| Gemini 3.5 Flash-Lite | 81.7 |
| GLM-5.3 | 81.7 |
| Opus 4.5 | 81.6 |
| Gemini 2.5 Pro Experimental 03-25 | 81.4 |
| Qwen3.5 Plus (2026-02-15) | 81.3 |
| GLM-5 | 81.2 |
| Qwen3.7-Max (source unspecified) | 81.2 |
| Sonnet 4 | 81.1 |
| Grok 4.7 | 81.0 |
| MiniMax M2.7 | 81.0 |
| MiniMax M3 | 81.0 |
| Grok 4.3 | 80.8 |
| Kimi K2.6 | 80.6 |
| GLM-5.1 | 80.6 |