API support by release

An environment serves a model only when its release ships an adapter for the API that model needs. This page lists what this release ships, generated from the adapters themselves, so a build cannot disagree with its own list.

What this release serves

Capability	API	Providers that reach it	Version	Default
Chat	anthropic:messages	anthropic	tested to 2023-06-01	✓
Chat	bedrock:converse	bedrock	n/a	✓
Chat	openai:chat_completions	azure-openai, openai, openai-compatible	tested to 2025-04-01-preview	✓
Chat	openai:responses	azure-openai, openai	≥ 2025-03-01-preview
Chat	vertex:generate_content	vertex	n/a	✓
Embeddings	openai:embeddings	azure-openai, openai	tested to 2025-04-01-preview	✓
Image generation	bfl:images	azure-openai	n/a
Image generation	mai:images	azure-openai	n/a
Image generation	openai:images	azure-openai, openai	tested to 2025-04-01-preview	✓
Image generation	vertex:generate_content	vertex	n/a	✓

A ✓ in the Default column marks the API a model resolves to when it declares none.

If an environment says it cannot serve a model

It is telling you the truth about ITSELF. An environment knows the adapters in its own build and nothing about releases that did not exist when it was built, so it never guesses which future release would help. That answer is this page.

Check the API named in the exception against the table above:

• Listed here, but not in that environment: the environment is on an older release. Upgrading it is the fix.
• Not listed here either: no release serves that API yet. The model cannot be used on this platform until an adapter ships, and the honest answer to “when?” is that it is not yet built.

Nothing about this is a fault in the model or the registry: the registry publishes the world as it is, and a client environment serves the part of it that its release implements.