What is OpenAI prioritizing under compute limits?

算力极限下，OpenAI 急着做什么？

Greg Brockman said OpenAI narrowed priorities under hard compute limits to two bets: a personal assistant and AI workers that solve hard user problems, and current compute cannot fully support both. The snippet says Sora resources were reduced while focus shifted to reasoning models, a unified AI layer, and the next base model Spud; it does not disclose the claimed compute budget, timeline, or model specs. The key point is not a B2B retreat but a compute-driven reprioritization.

Why it matters: HKR-H/K/R all pass: the compute-ceiling angle is strong, the piece adds concrete priority shifts, and OpenAI roadmap triage hits cost and dependency nerves. It stays at 80 because this is secondary reporting; spend, timing, and technical details are not disclosed.