GPT-6 Astra first day: code review stops inventing problems
A user says Astra feels about twice as smart as GPT-5.6 Sol, flagged exactly 3 real issues in a code review, and used only 1% of the weekly limit in an hour.
First impressions of GPT-6 Astra 1. A very terse model 2. It's the first model that doesn't invent problems when auditing code. I asked it to do a code review of a project. GPT-5.6 Sol would come back with 15 issues, half of which were pointless over-engineering and false claims. Astra pinpointed exactly 3 critical issues and gave an honest assessment of what had been done right 3. The limits are very generous. I've been working with it for about an hour and have used only 1% of my weekly limit. Hope they don't cut that 4. Pretty fast. On xhigh it feels a lot faster than GPT-5.6 Sol 5. It stopped introducing regressions into the project's code So far it looks about twice as smart as GPT-5.6 Sol