kontinent / status
Measured model availability
We call every served chat model every 30 minutes with the same prompt, through our own gateway. What you see here is the result of those calls — not a third-party report and not an estimate.
No probe calls recorded yet.
How we measure
The source is our own probe traffic only, never customer traffic. Every 30 minutes the same short prompt goes to every served chat model, plus one realistically long prompt once a day. Each bar is one day. “Available” is the share of successful calls over those same seven days; “Response” is the median of successful calls. A grey bar means we did not measure that day — not that the model was down. A provider takes the colour of its worst model.
When a model fails, the gateway tries the fallback chain behind it. The call still counts as available, because the customer did get an answer. The response time then includes the failed attempt. Where this happened it is noted under the model row, naming the model that actually answered. A green row carrying that note means: usable, but not answered by this provider.
This page describes the availability of individual models and providers. It is not a statement about the platform availability committed in § 7 of the Terms of Service — we expressly do not warrant any particular availability of individual models.