Embeddings

POST /v1/embeddings with model and input. OpenAI, Google, and Mistral embeddings are forwarded when that model is enabled and priced. Anthropic and xAI do not serve embeddings through Fuse.

input may be a string or an array of strings. There is no output cap. The reservation is the input token count times the embedding price. Cached input does not apply.

{
  "model": "text-embedding-3-small",
  "input": ["first chunk", "second chunk"]
}

The same Fuse key, ceiling, and allowlist apply. An embedding model you did not turn on is model_not_allowed. A chat model sent to this path is still priced as an embedding if that is the endpoint, so send embedding models here and chat models to chat completions.