PrinzBench
Professional Work · 2026-01-19
PrinzBench is a private legal-research benchmark that evaluates models on legal reasoning and obscure public-information search tasks. The public leaderboard reports a total score out of 99 points plus legal research and search subtotals.
Top models (higher is better)
| Model | Score |
|---|---|
| GPT-5.6 Sol Pro | 91.0 |
| GPT-5.5 Pro | 82.0 |
| GPT-5.4 Pro | 79.0 |
| GPT-5.5 | 74.0 |
| GPT-5.4 | 69.0 |
| Claude Fable 5 | 56.0 |
| GPT-5.2 | 52.0 |
| GPT-5.3-Codex | 52.0 |
| Gemini 3.1 Pro Preview | 50.0 |
| Kimi K3 | 47.0 |
| Grok 4.20 | 43.0 |
| Opus 4.8 | 42.0 |
| Gemini 3 Flash Preview | 36.0 |
| Gemini 3 Pro Preview | 35.0 |
| Kimi K2.5 | 35.0 |
| Muse Spark | 31.0 |
| GLM-5.2 | 30.0 |
| Grok 4.1 | 25.0 |
| Qwen3 Max (rolling alias) | 25.0 |
| Opus 4.7 | 25.0 |
| Grok 4 | 23.0 |
| DeepSeek-V4-Pro | 23.0 |
| Kimi K2 Thinking | 22.0 |
| Sonnet 4.5 | 20.0 |
| Opus 4.6 | 19.0 |
| Opus 4.5 | 14.0 |
| Sonnet 4.6 | 8.0 |