AIHubMix API Versioning and Deprecation Policy
How AIHubMix API versions, model retirement notices, Deprecation and Sunset headers guide client migrations.
Choose a versioned endpoint
Use the exact versioned path in the AIHubMix OpenAPI specification and the selected model guide. OpenAI-shaped APIs use /v1; Gemini uses its documented /gemini/v1beta paths; media APIs have their own versioned paths. A version in the URL identifies the request contract. It does not guarantee that every model or upstream parameter will remain available indefinitely. Check the selected model capabilities before migrating a client.
Model deprecation notices
API endpoint versions and model lifecycles are separate. For a model marked deprecated, the gateway uses Deprecation: @<unix-seconds> (RFC 9745) when a valid deprecation date is configured. A valid retirement date is communicated as Sunset: <HTTP-date> (RFC 8594). Link headers point to the model notice with rel="deprecation" and to a configured successor with rel="successor-version". These fields describe that model, not the retirement of every endpoint under /v1.
Plan a migration
Read the model retirement index and model-specific guide for the applicable dates and successor. The deprecated stage is a notice period; it does not itself reject requests. Once a model is retired, new inference requests for it are rejected; choose an available replacement and verify its supported parameters and response behavior. Header dates and successors depend on the published model configuration; a missing header is not a promise of unlimited availability.
Notice periods and endpoint changes
There is currently no universal advance-notice duration for all API endpoint or upstream model changes. Use the dates in the individual model notice rather than assuming a fixed 30-day or 90-day window. Contact support when your integration requires a contractual migration period. This page documents the existing lifecycle behavior; it does not announce deprecation of /v1 or a new public API version.