OpenAI Reliability
AI research and deployment company behind ChatGPT and the OpenAI API
Global and unknown-region reports remain visible. Scores use the selected incident history over the last 90 days.
Data notes · Provisional score
- Some incidents have incomplete timing evidence.
- Some incident timestamps are invalid or outside the reporting window.
- Some resolved incidents have missing or inconsistent fix-time measurements.
- Some incidents have missing or invalid communication measurements.
- Historical incident records need verification against the current source.
The score uses 102 of 103 recorded incidents with usable timing. Incomplete or conflicting intervals are excluded from the calculation and remain in incident history. These notes are reassessed as the underlying records are repaired.
Score Breakdown
An uptime proxy, not measured availability. Overlapping incidents are added together, including across products and regions.
53 of 102 included incidents have measured fix times.
Severity-weighted incident count. Providers cover different numbers of products and regions, so comparisons are approximate.
102 of 102 included incidents have update counts; 102 have acknowledgement times.