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 | 2249 |
| Gemini 4 Argon | 2239 |
| GPT-6.1 Sol Pro | 2189 |
| Claude Opus 5.5 | 2158 |
| GPT-6 Sol Pro | 2085 |
| Claude Opus 5 | 2026 |
| Claude Sonnet 5.5 | 2015 |
| GPT-5.6 Sol Pro | 2011 |
| Claude Fable 5.1 | 1945 |
| GPT-5.5 Pro | 1938 |
| Claude Fable 5 | 1885 |
| Muse Wonder | 1876 |
| Grok 4.7 | 1870 |
| GPT-5.5 | 1862 |
| DeepSeek V4.1 Flash | 1837 |
| Grok 4.6 | 1829 |
| Gemini 3.8 Flash | 1820 |
| GPT-5.4 Pro | 1809 |
| Opus 4.8 | 1800 |
| GLM-5.3 Flash | 1798 |
| Gemini 3.7 Flash | 1780 |
| GLM-5.3 | 1776 |
| GPT-6 Luna Pro | 1760 |
| Muse Spark 1.3 | 1760 |
| Kimi K3 | 1726 |
| GPT-5.6 Luna Pro | 1719 |
| Gemini 3.1 Pro Preview | 1691 |
| GPT-5.4 | 1681 |
| Gemini 3.5 Flash | 1667 |
| Sonnet 5 | 1640 |
| Gemini 3.6 Flash | 1634 |
| Opus 4.7 | 1623 |
| GPT-5.3-Codex | 1594 |
| Muse Spark 1.2 | 1582 |
| GLM-5.2 | 1577 |
| GPT-5.2 Pro | 1569 |
| Grok 4.5 | 1567 |
| GLM-5.1 | 1529 |
| DeepSeek-V4-Flash-0731 | 1512 |
| Opus 4.6 | 1502 |