Dev benchmarks Astra 6 vs 6.2 on 3D house modeling: less prompting now matches launch-day results

Dimillian · x · 2026-11-19

Developer Dimillian ran a small no-iteration benchmark comparing Astra 6 vs 6.2 on 3D house modeling, finding that 6.2 gets close to the results from his original Astra launch article with noticeably less prompting — a sign of improved instruction following. The OpenAI developer docs' architectural visualization demo serves as the reference case.

Original post →

More from Models

Models channel →