TechCrunch reported on 2 September that Astra uses a technique called recurrent depth. The model repeats a query internally instead of writing every reasoning step out in text. OpenAI says it kept the loops shallow, so Astra's visible reasoning should stay readable.