Embeddings

curl -X POST "$GATEWAY/v1/embeddings" \
  -H "Authorization: Bearer $API_KEY" \
  -H "Content-Type: application/json" \
  -d '{ "model": "embed", "input": ["red fox", "blue whale"], "encoding_format": "float" }'

Fields

model, input (a string, a string batch, pre-tokenized number[]/number[][], depending on the model's profile), encoding_format (float or base64), dimensions, user.

Every field is validated against the target model's embedding.create catalog profile before any upstream call: dimensions is only honored when the model declares supportsDimensions: true; encoding_format: "base64" is rejected unless the profile lists it; pre-tokenized input is rejected unless supportsTokenInput: true. A rejected combination returns 400 unsupported_parameter.

Provider routing

OpenAI, Azure OpenAI, and OpenAI-compatible deployments route to the native /embeddings endpoint. Google AI Studio routes to :embedContent (single input) or :batchEmbedContents (batch) — and, in this gateway, accepts text input and float vectors only, regardless of what a custom profile might otherwise allow. See Providers → Google AI Studio.

What to read next

• Model catalog → Embeddings — the full embedding.create schema.
• Creating deployments — registering an embeddings model per provider.