All models operational

Uptime is successful observations divided by total observations in the displayed period, including recorded requests, jobs and historical probes. A response from a substitute model does not count as success for the requested model.

Homepage and model comparisons show 30-day measurements. Status history shows the period marked on each timeline. These measurements describe past responses, not guaranteed future availability.

For eligible failures, automatic fallback tries another compatible route when available. A fallback can use another model. Applications should inspect the returned model and handle errors.