flâneur

Blueprint-Bench: Testing spatial intelligence in AI models | Andon Labs

andonlabs.com · 510 words · saved by 1 readers

How do AI models understand space? We test this by asking them to convert apartment photographs into accurate 2D floor plans. While photos are familiar training data, spatial reconstruction requires genuine intelligence.

Eval How do AI models understand space? We test this by asking them to convert apartment photographs into accurate 2D floor plans. While photos are familiar training data, spatial reconstruction requires genuine intelligence. Leaderboard Model Type Similarity Score (mean) 1 Human* Human 0.547 2 GPT-5 LLM 0.431 3 Gemini 2.5 Pro LLM 0.421 4 GPT-5-mini LLM 0.400 5 Grok-4 LLM 0.393 6 Codex CLI (GPT-5) Agent 0.388 7 Gemini 2.5 Flash LLM 0.362 8 Claude Code (Opus 4) Agent 0.355 9 GPT Image Image Model 0.300 10 Claude Opus 4 LLM 0.286 11 Random baseline…

saved by

related reading