OverblastAI gateway status
Checking…
Reading the last hour of requests.
Components
Each judged on its own requests over the last 24 hours.
Checking…
Where the time goes
Every request split into the part spent in Overblast and the part spent with the model provider.
Loading…
The last hour, minute by minute
The slowest 5% of each minute’s requests, end to end.
Loading…
Breakdowns
The last 24 hours. Times are median / slowest 5%.
Loading…
Slowest models (24h)
By the slowest 5% of their requests, end to end. Only models with at least five requests are ranked.
Loading…
What is measured
- Real traffic only: every model request to the AI gateway at /ai/v1, in the EU, US and global regions. Nothing here is a synthetic probe.
- Refused requests (no key, over a limit, out of balance) are counted in the requests, and never in the times, because they never reach a model provider.
- A failure is a request the model provider did not answer properly, or one we did not. Bad requests and refusals are not failures.
- “Overblast (ours)” is the time spent in Overblast before the model provider is called. The rest is the model provider. Model providers are not named on this page.
- Percentiles do not add up, so the total beside each bar is measured on its own and will not equal its three parts.
- All times are UTC. The data is updated every minute, and this page checks again every minute while it is open.