CadQueryEval — Hausdorff 95p
Code · 2026-01-10
Percentage of CadQueryEval tasks whose 95th-percentile bidirectional Hausdorff distance from the aligned reference geometry is at most 1.0 mm.
Top models (higher is better)
| Model | Score |
|---|---|
| GPT-5.6 Sol Pro | 96.0 |
| Gemini 3.1 Pro Preview | 92.0 |
| GPT-5.6 Luna Pro | 88.0 |
| GPT-5.6 Sol | 88.0 |
| Kimi K3 | 88.0 |
| GPT-5.6 Luna | 84.0 |
| Qwen3.7-Max | 84.0 |
| Claude Fable 5 | 80.0 |
| GPT-5.6 Terra Pro | 80.0 |
| Opus 4.6 | 76.0 |
| Opus 4.8 | 76.0 |
| Gemini 3.5 Flash | 76.0 |
| Sonnet 5 | 72.0 |
| GPT-5.5 | 72.0 |
| Grok 4.5 | 72.0 |
| Grok 4.3 | 68.0 |
| Muse Spark 1.1 | 68.0 |
| Opus 4.7 | 64.0 |
| Gemini 3 Pro Preview | 64.0 |
| GLM-5.1 | 64.0 |
| GPT-5.6 Terra | 64.0 |
| Kimi K2.6 | 64.0 |
| GPT-5 Mini | 60.0 |
| Qwen3.7-Plus | 60.0 |
| Kimi K2.5 | 56.0 |
| MiniMax M3 | 56.0 |
| GLM-5.2 | 52.0 |
| Hy3 preview | 52.0 |
| Opus 4.5 | 48.0 |
| Sonnet 4.5 | 48.0 |
| DeepSeek-V4-Pro | 48.0 |
| Gemini 3 Flash Preview | 48.0 |
| O1 | 48.0 |
| GPT-5.4 | 44.0 |
| Qwen3.6 Plus (2026-04-02) | 44.0 |
| Claude 3.5 Sonnet (Oct 2024) | 40.0 |
| DeepSeek-V4-Flash | 40.0 |
| Gemini 3.1 Flash-Lite | 40.0 |
| GPT-5 | 40.0 |
| GPT-5.2 | 40.0 |