LIVE STATUS
Is model inference working right now?
This page shows what our own traffic to the model catalogue is experiencing, measured on real requests, next to what the services behind the accounts you can connect yourself say about themselves.
Measured by our own traffic
From the circuit breaker on real requests we made in the last minutes — not a self-report. A provider we have not called recently shows as idle rather than a green light we cannot justify.
Services behind your own connected accounts
Claude, Codex and Gemini sign-ins run on these. Straight from their own public status pages. Their word, not ours — these often lag what users are actually seeing.
What happens to your request during an outage
A failing model is retried once. If it keeps failing, its circuit breaker opens and we stop sending it work until a probe succeeds. Your own fallback rules in the settings decide which model answers instead; the chat says which one did.