Codex reports session hang incident and rate-limit reset after fix
OpenAI acknowledged a Codex session hang that left some requests unresponsive, later said the issue had been stable for hours, and promised a rate-limit reset. Teams relying on Codex should re-check long runs and confirm quota restoration after the incident.
TL;DR
- OpenAI said some Codex users were seeing sessions “not responsive after sending a request,” with the team investigating a hang affecting request handling in active workflows, according to the incident post.
- A user reply added that Codex had “been running smoothly today” after seeing one hang the prior evening, suggesting the issue was intermittent rather than a universal outage, as described in the user report.
- OpenAI later said the Codex issue was “fully resolved and stable for the last couple of hours,” according to the resolution update.
- The same resolution update said rate limits would be reset after the fix, which matters for teams that burned quota during stalled or retried runs.
What happened
OpenAI’s incident post described the failure mode narrowly: Codex could hang after a request was sent, leaving the session unresponsive while the team investigated. That points to a runtime or session-state problem rather than a model deprecation, pricing change, or planned product update.
A supporting user report gives one operator-level signal about impact. The user said they had seen “one last evening,” but that Codex had “been running smoothly today,” which suggests at least some sessions recovered before the public all-clear and that the incident may have hit users unevenly across time windows.
What changed after the fix
OpenAI’s resolution update said the issue had been “fully resolved” and had remained stable for “the last couple of hours” before the announcement. That is the key operational change: the company moved from active investigation to a stability claim, with no indication in the provided evidence of a remaining degraded state.
The same resolution update also promised a rate-limit reset “in a bit.” For engineers using Codex in long interactive sessions or repeated retries, that matters as much as the fix itself, because quota consumed during hangs can block follow-on work even after service recovers. The evidence here does not specify whether resets were global, automatic, or already completed at posting time; it only confirms the reset was planned after stability had been restored.