November 6, 2026
Astra writes worse code for safety research: gap reaches 6 percentage points

OpenAI found that Astra performs worse on essentially identical tasks when they are framed as safety research.
Documentation
Journal
Frontend
Backend
Admin
CI/CD
Loading the page.
Loading the page.
November 6, 2026

OpenAI found that Astra performs worse on essentially identical tasks when they are framed as safety research.
who it's about
about the same entities
One email a week - only what changes how you work. Subscribe