MineBench — Bradley-Terry Rating
Games
MineBench ranks Minecraft-style voxel builds from blind human pairwise votes using a global Bradley-Terry model on a 1500-centered Elo scale. Ties contribute half a point; both-bad votes are excluded from the skill fit. This rating replaces the earlier conservative Glicko score and is kept as a separate protocol.
Top models (higher is better)
| Model | Score |
|---|---|
| GPT-6 Astra Pro | 2262 |
| Claude Opus 5.5 | 2150 |
| Claude Opus 5 | 2063 |
| GPT-5.6 Sol Pro | 2057 |
| GPT-6 Sol Pro | 2057 |
| Claude Fable 5.1 | 1984 |
| GPT-5.5 Pro | 1982 |
| Claude Fable 5 | 1925 |
| GPT-5.5 | 1904 |
| DeepSeek V4.1 Flash | 1887 |
| Grok 4.6 | 1884 |
| Gemini 3.8 Flash | 1871 |
| GPT-5.4 Pro | 1850 |
| GLM-5.3 Flash | 1843 |
| Opus 4.8 | 1841 |
| Gemini 3.7 Flash | 1828 |
| GLM-5.3 | 1818 |
| Muse Spark 1.3 | 1802 |
| Kimi K3 | 1769 |
| GPT-5.6 Luna Pro | 1762 |
| GPT-6 Luna Pro | 1761 |
| Gemini 3.1 Pro Preview | 1732 |
| GPT-5.4 | 1721 |
| Gemini 3.5 Flash | 1710 |
| Gemini 3.6 Flash | 1685 |
| Sonnet 5 | 1681 |
| Opus 4.7 | 1664 |
| GPT-5.3-Codex | 1634 |
| Muse Spark 1.2 | 1624 |
| GLM-5.2 | 1620 |
| GPT-5.2 Pro | 1608 |
| Grok 4.5 | 1605 |
| GLM-5.1 | 1571 |
| DeepSeek-V4-Flash-0731 | 1556 |
| Opus 4.6 | 1541 |
| GPT-5.2 | 1525 |
| DeepSeek-V4-Pro | 1509 |
| Kimi K2.6 | 1494 |
| GPT-5.4 Mini | 1454 |
| Gemini 3 Pro Preview | 1442 |