Should OpenAI Keep GPT-6 Astra's Successor Internal? The Compute Gap Problem

haider1 · x · 2026-09-26

The author suggests OpenAI should keep Astra's successor internal for research and distill cheaper models from it. With astra 6.1 much better than 6.0, the compute situation worsens: OpenAI needs the next model to be stronger than Astra, cheaper per task, and perhaps 5-10x cheaper on compute internally. NVIDIA Rubin will help but takes months to roll out, while user demand can explode in weeks — a gap that isn't sustainable. (GPT-6 Astra is unverified.)

Original post →

More from Models

Models channel →