Embeddings

POST https://models.mixlayer.ai/v1/embeddings

Create vector embeddings from text or token inputs using an embedding model. The endpoint follows the OpenAI-compatible embeddings response shape and is currently in beta.

⚠️

Embeddings support is in beta. Use an embedding model authorized for your API key; chat/generation models are rejected on this endpoint.

Authentication

Every request requires a Bearer token in the Authorization header. Create one in the Mixlayer console.

Authorization: Bearer $MIXLAYER_API_KEY

Minimal request

curl https://models.mixlayer.ai/v1/embeddings \
  -H "Authorization: Bearer $MIXLAYER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "$MIXLAYER_EMBEDDING_MODEL",
    "input": "The food was delicious and the service was fast."
  }'

Request parameters

This page is the canonical reference for supported embeddings parameters.

Parameter	Type	Required	Description
`model`	string	Yes	Embedding model identifier. The model must be available to your API key and marked as an embedding model.
`input`	string, string array, token array, or array of token arrays	Yes	Input to embed. Text strings must be non-empty. Arrays must contain 1-2048 items. Token inputs must be non-empty and fit the model context window.
`dimensions`	integer	No	Requested output vector size. Must be greater than `0` when provided. See Embedding model-specific behavior below.
`encoding_format`	`"float"` or `"base64"`	No	Format for returned embedding vectors. Defaults to `"float"`.
`input_type`	string	No	Optional model-specific embedding task hint. See Embedding model-specific behavior below.
`normalize`	boolean	No	Requests normalized vectors when supported by the model server.
`truncate`	boolean	No	Requests truncation behavior when supported by the model server.

Embedding model-specific behavior

Qwen3 Embedding 8B

Supported dimensions values: integers from 32 to 4096.

Supported input_type values:

Value	Use for
`query`	Search queries, questions, or other text used to retrieve matching documents.
`document`	Documents, passages, or other corpus text being retrieved.

Omit input_type when you do not want a model-specific prompt applied.

Input formats

Text input:

{
  "model": "$MIXLAYER_EMBEDDING_MODEL",
  "input": "Embed this text"
}

Batched text input:

{
  "model": "$MIXLAYER_EMBEDDING_MODEL",
  "input": ["First document", "Second document"]
}

Token input:

{
  "model": "$MIXLAYER_EMBEDDING_MODEL",
  "input": [101, 7592, 102]
}

Batched token input:

{
  "model": "$MIXLAYER_EMBEDDING_MODEL",
  "input": [[101, 7592, 102], [101, 2088, 102]]
}

Response

A successful response returns a list of embedding objects:

{
  "object": "list",
  "data": [
    {
      "object": "embedding",
      "embedding": [0.0023064255, -0.009327292, 0.015797347],
      "index": 0
    }
  ],
  "model": "$MIXLAYER_EMBEDDING_MODEL",
  "usage": {
    "prompt_tokens": 8,
    "total_tokens": 8
  }
}

When encoding_format is "base64", each embedding value is a base64 string instead of an array of floats.

Errors

Errors use the OpenAI error envelope:

{
  "error": {
    "message": "input strings must not be empty",
    "type": "invalid_request_error",
    "code": "invalid_input"
  }
}

Common validation errors include:

Code	Cause
`invalid_input`	Empty input, empty text, too many batch items, or token input longer than the model context window.
`invalid_dimensions`	`dimensions` was `0`.
`invalid_encoding_format`	`encoding_format` was not `"float"` or `"base64"`.
`model_kind_mismatch`	The requested model is not an embedding model.

Responses Tool Calling