Dev's Deep Dive: GPT-6.2 Astra Improved but Still 'Directionally Very Bad' at Debugging

kimmonismus · x · 2026-11-18

Developer theo spent hours stress-testing OpenAI's GPT-6.2 Astra and concluded it's better than his earlier tests but still 'directionally, very bad.' His recurring test case — 'why is creating a worktree so slow?' — exposed the model: the actual cause was a notoriously slow virtual disk on his VPS, but Astra responded with gibberish when asked directly for the root cause. kimmonismus amplified the thread, adding that even OpenAI employees he's spoken to privately concede Astra kinda sucks.

Related event: Developer Tests GPT-6.2 Astra: Better, But Still on the Wrong Track(2 posts)→

Original post →

More from Models

Models channel →