Live status
Which models have live capacity right now. One cell per hour over the last 24 hours. Green is healthy, amber is degraded, red is failing, grey is idle and was never scored as up. Prices live in the model directory. We never name the seller or the channel.
—
Live now
—
Listed, not serving
—
With supply
Reading the supply board...
Models with live or listed supply
Reading the supply board...
How we measure
Availability is the share of settled requests that completed without an error. A request counts as an error when the engine reports a failure or the request ends with a 4xx/5xx status. Client-side outcomes we cannot see — a caller that disconnected, or a request refused before it reached the engine — are excluded from the per-model figure rather than counted as downtime.
Each cell is one settled hour of our own receipts in the selected window. Green is under 1% errors, amber is under 10%, red is 10% or more, and grey is idle — an hour with no settled request, shown as idle and never scored as up. A model that barely served shows mostly grey cells instead of a painted-up average.
The headline rate counts settled requests in the window. Auto-routed models are rated the same way, from the requests that named the model. When there are too few requests to judge, the window reads No traffic rather than 100%.
Time to first token and throughput stay Not measured until the engine timing is verified against our own request timeline; the sample counts show how much evidence exists. The board names no seller and no channel, and it carries no prices — those live in the model catalog.