Read coverage, concurrency and reliability measurements

Model count, concurrency, success rate and reliability require different measurement definitions. Public catalogs can count unique exact IDs, but not prove all models tested. Health-check uptime, request acceptance and task completion must remain separate. Consistent definitions help developers evaluate access by task, cost and response needs.

Coverage

Report date, unique exact IDs, verifiable providers/families and execution-tested scope separately.

Concurrency and latency

Specify model, input/output length, concurrency or request rate, duration, sample size, failure definition, first response and completion latency.

Success and reliability

Count synchronous text business-response success and asynchronous media final completion separately. Report health checks with observation windows and incident records.

Apply these measurements

First confirm exact model IDs, pricing units and request formats, then compare response behavior under a defined observation window and workload. The API service page explains direct access; Model Plaza lists current IDs and prices. Actual permissions follow your account settings.