Benchmarks · reasoning

Blueprint-Bench 2

Spatial reasoning: reconstruct 2D floor plans from ~20 apartment photos, scored on room connectivity and scale (random baseline normalized to 0).

30 models · 30 results · data as of 2026-10-29 · andonlabs.com ↗

All results

30 rows

#	Model	Effort	Blueprint-Bench 2
1	Gemini 4 Carbon	max	58.6%
2	Gemini 4 Argon	max	54.4%
3	Opus 5.5	max	51.2%
4	GPT-6 Astra	max	49.7%
5	GPT-6.1 Terra	max	42.1%
6	Fable 5.1	max	41.9%
7	Sonnet 5.5	max	39.3%
8	Fable 5	max	38.6%
9	Gemini 3.8 Flash	max	38.6%
10	GPT-6 Sol	max	36.9%
11	GPT-5.5	xhigh	36.2%
12	Gemini 3.5 Flash	xhigh	33.6%
13	GPT-5.6 Sol	max	33.6%
14	Grok 4.6	max	33.2%
15	Haiku 5.5	max	33.0%
16	Grok 4.7	max	32.5%
17	Gemini 3.6 Flash	xhigh	31.2%
18	GPT-6 Luna	max	31.2%
19	GPT-5.6 Terra	max	30.8%
20	Opus 5	max	30.4%
21	Kimi K3	max	29.5%
22	Grok 4.5	max	27.3%
23	GPT-5.4	xhigh	27.1%
24	Gemini 3.1 Pro	high	26.5%
25	Sonnet 5	max	24.9%
26	Opus 4.7	max	24.5%
27	GPT-5.6 Luna	max	22.6%
28	Opus 4.8	max	14.5%
29	Sonnet 4.6	max	6.7%
30	Kimi K2.6	max	3.9%