VoxelBench (text)

Models build 3D voxel scenes from text prompts; humans vote head-to-head on the renders. Glicko-2 rating.

54 models · 54 results · data as of 2026-11-06 · voxelbench.ai ↗

All results

54 rows

#	Model	Effort	VoxelBench (text)
1	GPT-6 Astra	max	2725 ±261
2	Opus 5.5	max	2606 ±125
3	GPT-6.1 Sol	max	2550 ±229
4	GPT-6 Sol	max	2394 ±120
5	Opus 5	max	2313 ±114
6	GPT-6.1 Terra	max	2312 ±74
7	GPT-5.6 Sol	max	2296 ±108
8	Kimi K3.1	max	2216 ±84
9	Gemini 4 Argon	max	2214 ±108
10	Fable 5	max	2098 ±112
11	Grok 4.6	xhigh	2050 ±110
12	DeepSeek V4.1 Flash 1017	max	2049 ±108
13	GPT-5.5 Pro	xhigh	1945 ±112
14	Kimi K3	max	1939 ±106
15	GPT-5.5	xhigh	1925 ±108
16	GPT-5.6 Terra	max	1896 ±104
17	DeepSeek V4.1 Flash	max	1861 ±114
18	GPT-5.6 Luna	max	1857 ±110
19	Gemini 3.7 Flash	max	1830 ±104
20	GPT-6 Luna	max	1817 ±114
21	MiniMax M3.1	max	1808 ±108
22	Gemini 3.8 Flash	max	1799 ±110
23	Qwen 3.8 Max	max	1790 ±104
24	Gemini 3.6 Flash	xhigh	1775 ±106
25	GLM 5.3	max	1773 ±106
26	Grok 4.5	max	1737 ±106
27	GLM 5.3 Flash	max	1735 ±106
28	Haiku 5.5	max	1697 ±110
29	Gemini 3.5 Flash	xhigh	1635 ±110
30	Opus 4.8	max	1628 ±110