Blueprint-Bench 2
Multimodal · 2026-05-04
Blueprint-Bench 2 evaluates spatial reasoning by asking persistent agents to reconstruct floor plans for 50 apartments from approximately 20 photographs each. Published scores normalize connectivity similarity so the random baseline is zero and perfect reconstruction is one.
Top models (higher is better)
| Model | Score |
|---|---|
| Gemini 4 Carbon | 0.6 |
| Human (Blueprint-Bench 2) | 0.6 |
| Gemini 4 Argon | 0.5 |
| Claude Opus 5.5 | 0.5 |
| GPT-6 Astra | 0.5 |
| Claude Fable 5.1 | 0.4 |
| Claude Sonnet 5.5 | 0.4 |
| Claude Fable 5 | 0.4 |
| Gemini 3.8 Flash | 0.4 |
| GPT-6 Sol | 0.4 |
| GPT-5.5 | 0.4 |
| Gemini 3.5 Flash | 0.3 |
| GPT-5.6 Sol | 0.3 |
| Grok 4.6 | 0.3 |
| Claude Haiku 5.5 | 0.3 |
| Grok 4.7 | 0.3 |
| Gemini 3.6 Flash | 0.3 |
| GPT-6 Luna | 0.3 |
| GPT-5.6 Terra | 0.3 |
| Claude Opus 5 | 0.3 |
| Grok 4.5 | 0.3 |
| GPT-5.4 | 0.3 |
| Gemini 3.1 Pro Preview | 0.3 |
| Sonnet 5 | 0.2 |
| Opus 4.7 | 0.2 |
| GPT-5.6 Luna | 0.2 |
| Opus 4.8 | 0.1 |
| Sonnet 4.6 | 0.1 |
| Kimi K2.6 | 0.0 |
| Haiku 4.5 | 0.0 |
| Gemini 3 Flash Preview | 0.0 |
| Gemini Robotics-ER 1.6 | 0.0 |
| Grok 4.20 | 0.0 |
| Grok 4.3 | 0.0 |