Astra tops Blueprint-Bench 2, the spatial-reasoning benchmark turning apartment photos into floor plans
Wonderful_Buffalo_32 · reddit · 2026-09-08
A Reddit post reports Astra scoring highest on Blueprint-Bench 2, a spatial-reasoning benchmark where AI agents convert apartment photographs into accurate 2D floor plans.
Each agent processes 50 apartments sequentially, examining 20 interior photos per apartment and generating floor plans with room layouts, connections, and relative sizes. A persistent notepad lets agents carry insights across apartments for cross-apartment learning and iterative strategy refinement.
Related event: GPT-6 Astra Tops Blueprint-Bench 2 in Spatial Reasoning(2 posts)→
More from Models
- Every tested model reinforced user delusions in mental health scenarios, with safety only in ~40% of turns — alex_verem · 2026-09-08
- LLM Museum: historical benchmarks and infographics of frontier models since GPT-4 — delusion54 · 2026-09-08
- GPT-6 Astra generates 13,038 annotations across 81 warehouse CCTV frames — luisdans · 2026-09-08
- After 3 days of testing, GPT-6 Astra looks overhyped: overfit and messy code — ivan_bezdomny · 2026-09-08
- Dev wants ChatGPT desktop but multi-account limits force him back to the TUI — kevinkern · 2026-09-08
- Gemini 'lied' about image quota: AI Pro upgrade turns out web-app only — Morpegom · 2026-09-08