AI Incidents
Back to incident register
Source reviewedOperational incidentLow

OpenAI API experiences elevated error rates on March 4

A batched execution of queued capacity actions reduced model capacity for about 30 minutes and increased API errors and latency.

First observed
Mar 5, 2026, 1:26:00 AM
Disclosed
Mar 5, 2026, 1:26:00 AM
Status
Resolved
Confidence
99%
Organization
OpenAI
Last reviewed
Sep 1, 2026

Observed

Facts supported by sources

  • API success rates fell by varying amounts across models; some Codex traffic was also affected.
  • The disruption lasted approximately 30 minutes.
  • The UTC time was converted from the PST times stated in the postmortem.

Assessment

Interpretation, not observation

No additional interpretation recorded.

Impact

Impact

Elevated latency and failed API requests across multiple models.

Response

Response

OpenAI manually reallocated capacity and added safeguards against batched scheduler actions.

Methodology: Facts, interpretations, and uncertainty are kept separate. Confidence describes the strength of the evidence, not a probability estimate.