PrinzBench Legal Research
Professional Work · 2026-01-19
The legal-research subscore of PrinzBench covers 25 legal research questions, each asked three times, for a maximum of 75 points. It measures a model's ability to locate and analyze relevant public legal authorities.
Top models (higher is better)
| Model | Score |
|---|---|
| GPT-5.6 Sol Pro | 67.0 |
| GPT-5.5 Pro | 63.0 |
| GPT-5.4 Pro | 59.0 |
| GPT-5.5 | 56.0 |
| GPT-5.4 | 50.0 |
| Claude Fable 5 | 49.0 |
| GPT-5.2 | 41.0 |
| GPT-5.3-Codex | 41.0 |
| Gemini 3.1 Pro Preview | 41.0 |
| Kimi K3 | 36.0 |
| Opus 4.8 | 35.0 |
| Grok 4.20 | 34.0 |
| Gemini 3 Flash Preview | 29.0 |
| Gemini 3 Pro Preview | 29.0 |
| Kimi K2.5 | 28.0 |
| GLM-5.2 | 25.0 |
| Muse Spark | 24.0 |
| Opus 4.7 | 21.0 |
| Sonnet 4.5 | 20.0 |
| Grok 4.1 | 19.0 |
| Qwen3 Max (rolling alias) | 18.0 |
| Kimi K2 Thinking | 18.0 |
| Opus 4.6 | 18.0 |
| DeepSeek-V4-Pro | 17.0 |
| Grok 4 | 16.0 |
| Opus 4.5 | 14.0 |
| Sonnet 4.6 | 8.0 |