Thibault Sottiaux, engineering lead on OpenAI’s Codex team, posted an informal poll on May 24 asking whether Codex should gain batch compute support under the label “/slow mode” — a mechanism that would process compute-intensive coding tasks on a delayed or grouped basis. The post drew over 114,000 views and 810 replies, with 92% of sentiment-tracked responses in favour. Developers cited overnight task queues and long-running /goal workflows as primary use cases, with Matthew Berman noting he would willingly accept slower processing in exchange for lower cost.
The proposal echoes OpenAI’s existing Batch API, which offers 50% cost reduction in exchange for up to 24-hour turnaround on inference requests. Applying a similar model to Codex would let developers kick off large refactors or multi-step automations without consuming real-time quota. A minority of replies argued that existing reliability issues should be addressed before new features are added. No timeline or formal commitment was given; the post reads as an early-stage feature gut-check with the user community.