Astra launches: saturated evals, better alignment, but code slop and over-confirmation remain

ren_hongyu · x · 2026-09-04

Team member @yanndubs announces Astra's release. Highlights: much smarter (evals saturating), more aligned and trustworthy, and stronger computer-use ability — it even built a house in Blender. Known issues: too much code slop and excessive confirmation-asking (overcorrected caution), both slated for the next fix.

Related event: OpenAI ships GPT-6 Astra: team says smarter and more aligned, but code slop persists(6 posts)→

Original post →

More from Models

Models channel →