PrinzBench Search
Search · 2026-01-19
The search subscore of PrinzBench covers 8 needle-in-the-haystack search questions, each asked three times, for a maximum of 24 points. It measures a model's ability to find obscure publicly available information online.
Top models (higher is better)
| Model | Score |
|---|---|
| GPT-5.6 Sol Pro | 24.0 |
| GPT-5.4 Pro | 20.0 |
| GPT-5.5 Pro | 19.0 |
| GPT-5.4 | 19.0 |
| GPT-5.5 | 18.0 |
| GPT-5.2 | 11.0 |
| GPT-5.3-Codex | 11.0 |
| Kimi K3 | 11.0 |
| Gemini 3.1 Pro Preview | 9.0 |
| Grok 4.20 | 9.0 |
| Claude Fable 5 | 7.0 |
| Opus 4.8 | 7.0 |
| Gemini 3 Flash Preview | 7.0 |
| Kimi K2.5 | 7.0 |
| Muse Spark | 7.0 |
| Qwen3 Max (rolling alias) | 7.0 |
| Grok 4 | 7.0 |
| Gemini 3 Pro Preview | 6.0 |
| Grok 4.1 | 6.0 |
| DeepSeek-V4-Pro | 6.0 |
| GLM-5.2 | 5.0 |
| Opus 4.7 | 4.0 |
| Kimi K2 Thinking | 4.0 |
| Opus 4.6 | 1.0 |
| Sonnet 4.5 | 0.0 |
| Opus 4.5 | 0.0 |
| Sonnet 4.6 | 0.0 |