The same model has different IDs by host
A first-party API, Azure deployment, Bedrock model card, and aggregator may use different namespaces for the same family. The model name in a launch announcement is not enough to configure a backend.
Model IDs are operational configuration, not permanent facts. A reliable routing workflow combines provider model APIs, deployment availability, and an explicit review date.
By Sarah Reid · ·
A first-party API, Azure deployment, Bedrock model card, and aggregator may use different namespaces for the same family. The model name in a launch announcement is not enough to configure a backend.
AI-FW's model reference provides representative IDs and base URLs, but the provider's own catalog remains authoritative. Use the live list for account-specific availability, regional deployments, preview access, and retirement status.
In Model Inventory, register the provider URL, model ID, authentication, API style, and routing behavior. Test the selected deployment before enabling strict model enforcement for production traffic.
Model routing does not replace prompt inspection, response inspection, identity, or audit. The gateway keeps those controls in the path while the provider catalog changes behind it.