Roybench availability

Provider and model uptime

Compare Roybench-observed model availability with provider-published incident history. These are separate evidence sources and are never blended into one score.

Loading both uptime sources…
Source
Waiting for data
Provider average
Rows with available evidence
Window
7 days
Waiting for data

Uptime by provider-qualified route

Success / explicit success + failure observations

Bar chart of uptime percentages. Use Left and Right arrow keys to inspect bars.

UptimeObservation coverage
Provider uptime
ProviderEvidenceUptimeCoverage / incidentsCurrent source statusSource freshness
Provider-qualified model uptime
ModelRouteProviderUptimeEvidence scopeCoverage / incidents

What the numbers mean

Roybench observations: successful checks divided by explicit success plus failure checks in the selected window. Each provider percentage is weighted across its provider-qualified model checks. Collection gaps and administrative pauses are excluded; coverage compares eligible checks with expected half-hour buckets.

Official status pages: incident-free time in the selected window after unioning overlapping incident intervals. Official status is provider- or service-level; model rows inherit the provider percentage and do not imply model-specific official evidence. Public histories can be incomplete, and Google reports only broad severe Vertex Gemini incidents. Source failures remain stale or unavailable; they are never inferred operational.

Unsupported source: Z.ai has no verified public official status feed, so its official percentage remains unknown. Roybench observations remain a separately labeled signal.