All providers reachable
Probed directly from our servers when you load this page, at most once every five minutes. Nothing on this page is a stored marketing figure.
We publish two separate signals and never blend them. A probe asks the provider for its model list: cheap, immediate, and it answers “is anyone home”. Real traffic success is the share of routed requests that actually completed, which is slower to move but is what your users experienced. A provider can pass every probe while failing every completion, and averaging the two would hide exactly that.
We publish no uptime SLA. We have not been running long enough for one to mean anything, and a number we cannot evidence is worth less than no number at all. What is below is the measurement itself, with its gaps left in: every ribbon is a row of time slots, and a slot we did not observe is drawn empty rather than green. A rate appears once a lane holds 10 probes or 20 requests, and not before.
- Succeeded
- Failed on a degraded day
- Failed on a bad day (< 50%)
- Nothing measured yet
The ring is reachability probes, measured across 100% of the last 30 days a share of what we watched, not of the month. We probe when someone loads this page, so the gaps are ours.
The ring is reachability probes, measured across 100% of the last 30 days a share of what we watched, not of the month. We probe when someone loads this page, so the gaps are ours.
The ring is reachability probes, measured across 100% of the last 30 days a share of what we watched, not of the month. We probe when someone loads this page, so the gaps are ours.
The ring is reachability probes, measured across 100% of the last 30 days a share of what we watched, not of the month. We probe when someone loads this page, so the gaps are ours.
The ring is reachability probes, measured across 100% of the last 30 days a share of what we watched, not of the month. We probe when someone loads this page, so the gaps are ours.
The ring is reachability probes, measured across 100% of the last 30 days a share of what we watched, not of the month. We probe when someone loads this page, so the gaps are ours.
What happens when one of these is down
Models served by more than one provider fail over automatically: a 429 or 5xx moves the request to the next route before the first token is sent. Models with a single provider fail with a 502 or 504 naming that provider. We do not silently substitute a different model, because a quietly different answer is much harder to notice than an error.
| Signal | Refreshed | Window |
|---|---|---|
| Reachability probe | On page load, at most every 5 minutes | Latest result |
| Probe success rate | Same | 30 days, shown at 10+ samples |
| Real traffic success | Continuously | 30 days, shown at 20+ requests |
| Credit refusal | When a provider refuses a real request for credit | Marked on the ribbon, never averaged into a rate |