OpenAI's Astra uses 'recurrent depth' reasoning, obscuring its thinking process

JacquesThibs · x · 2026-09-05

Per The Information, OpenAI's Astra model uses a new reasoning approach called "recurrent depth." It can help with cost and performance, but researchers worry it obscures the model's thinking process, making monitoring and safety research harder. The resharing author argues limited transparency is making safety research increasingly difficult.

Original post →

More from Models

Models channel →