Benchmarks / 3D
threejseval
Coding models build a Three.js scene from the same prompt; humans vote blind on pairs of finished scenes. Value is Bradley-Terry Elo (base 1500), the mean across prompts.
threejseval
Higher is better · best effort per model
full table below
| # | Model | Effort | Score |
|---|---|---|---|
| 1 | Opus 5.5 | high | 2071 |
| 2 | GPT-6.1 Sol | high | 1988 |
| 3 | GPT-6 Astra | medium | 1884 |
| 4 | Fable 5.1 | high | 1866 |
| 5 | Sonnet 5.5 | high | 1837 |
| 6 | Gemini 4 | high | 1796 |
| 7 | Opus 5 | medium | 1713 |
| 8 | DeepSeek V4.1 Flash | max | 1662 |
| 9 | GPT-6 Sol | high | 1631 |
| 10 | Grok 4.7 | xhigh | 1625 |
| 11 | Gemini 3.8 Flash | high | 1571 |
| 12 | Kimi K3 | high | 1520 |
| 13 | Fable 5 | medium | 1520 |
| 14 | Gemini 3.7 Flash | high | 1473 |
| 15 | Grok 4.6 | high | 1461 |
| 16 | GPT-5.6 Terra | max | 1440 |
| 17 | Muse Spark 1.3 | max | 1439 |
| 18 | Haiku 5.5 | high | 1424 |
| 19 | GPT-5.6 Sol | medium | 1379 |
| 20 | GPT-6 Luna | high | 1375 |
| 21 | Gemini 3.6 Flash | high | 1348 |
| 22 | GPT-5.6 Luna | high | 1280 |
| 23 | Sonnet 5 | medium | 1212 |
| 24 | GPT-5.3 Codex | high | 1160 |
| 25 | Opus 4.5 | high | 1149 |
| 26 | Composer 2.5 | none | 1087 |
| 27 | Mistral Large 4 | high | 1052 |