Methodology

Every outage above is a period our probes measured OpenAI as down from enough of the regions we check it from to count as a real outage, not a single unreliable reading (see how we tell the difference). Measured downtime is the wall-clock time inside the window an outage's regions were failing; regions failing together still count as one outage, not one per region.

OpenAI's claim at the time comes from their own public status feed, polled on our own schedule, and is shown separately from what we measured. We never change our own reading based on what a vendor claims, and their claim never becomes our reading.

This is one external observation from RealUptime's own regions, not OpenAI's own SLA measurement, which is usually defined against their own instrumentation and may cover a different scope than what we probe. The credit estimate above is a starting point for a conversation with OpenAI, not a legal determination, and RealUptime is not counsel.