About ModelBank & methodology

ModelBank answers two questions about AI: is it up right now, and is it the model it claims to be? We check AI services with real API calls every 2 minutes and offer a free model authenticity check; prices and deprecations are tracked too. ModelBank is independent and not affiliated with any AI vendor.

Status monitoring

Every 2 minutes our monitoring network checks each service in two ways: (1) it reads the vendor’s official status page (when a machine-readable one exists), and (2) it sends an unauthenticated request to the vendor’s public API endpoint from our global monitoring nodes. Any HTTP response below 500 (including 401 “missing API key”) means the API front door is reachable. Timeouts, network errors and 5xx responses are retried once and then counted as a failure. A service is marked Outage when our probe fails or the vendor declares a major/critical incident, and Degraded for minor incidents, maintenance or response times above 5 seconds. Uptime = (operational checks + ½ degraded checks) / all checks.

Limitations: the probe confirms the API is reachable, not that every model returns correct completions. Model-level errors are reflected via official incidents and per-provider uptime.

Prices & models

New models are detected by the Model Radar, our proprietary monitoring system that watches public, official vendor channels around the clock. Prices and model metadata are refreshed every 10 minutes and cross-checked across multiple independent sources before publication. Prices are shown in USD per 1 million tokens. “Blended” price assumes 3 input tokens per output token. Every change is stored, so price history starts on the day tracking began. Benchmark indices are published third-party evaluations.

Contact

Corrections, data partnerships or listing a service: [email protected]