OpenAI Embeddings

The OpenAI Embeddings API converts one or more inputs into vector representations for semantic search, clustering, recommendations, and vector retrieval. Input limits, output dimensions, and optional parameters depend on the selected embedding model.

POST/v1/embeddings

Create vector representations for semantic search, clustering, recommendations, and other retrieval workflows. The API accepts a single text input, a batch of text inputs, a token array, or a batch of token arrays.

Input limits and output dimensions depend on the selected model. For example, text-embedding-ada-002 returns 1,536-dimensional vectors and accepts up to 8,192 tokens; other models may use different limits.

Authorization

bearerAuth

AuthorizationBearer <token>

Use Bearer your-api-key in the Authorization Header.

In: header

Request Body

Choose a text structure based on the number of inputs. Both views use the same Embeddings endpoint.

application/json

Single text

Convert one non-empty string into an embedding vector for single-item tests or real-time queries.

model: An embedding model ID visible to the current API key.
input: A single non-empty string to convert into a vector.
encoding_format: Vector encoding format. The default is float.

application/json

{
  "model": "MODEL_ID",
  "input": "The food was delicious and the waiter was friendly.",
  "encoding_format": "float"
}

model*string

The model ID to use. Please enter the Embedding model ID available for the current account.

Length1 <= length

input*|array<>|array<integer>|array<array<integer>>

Input to be converted into a vector. Single text, text array, integer array of tokens, or batch token arrays can be passed. Inputs cannot be empty; Batch quantity, total number of tokens, and upper limit per token depend on the selected model

encoding_format?string

Returns the encoding format of the embedded vector.

Default"float"

Value in

"float"
"base64"

dimensions?integer

The desired output vector dimension. Only some newer Embedding models support this parameter.

Range1 <= value

user?string

Unique identifier for the end user, used to help monitor and detect abuse.

Length1 <= length

Token arrays and batched token arrays remain available in the complete field reference. Input count, length, and vector dimensions depend on the selected model limits.

Response Body

`application/json`

curl -X POST "https://example.com/v1/embeddings" \  -H "Content-Type: application/json" \  -d '{    "model": "MODEL_ID",    "input": "The food was delicious and the waiter was friendly.",    "encoding_format": "float"  }'

{
  "object": "list",
  "data": [
    {
      "object": "embedding",
      "embedding": [
        0.0023064255,
        -0.009327292,
        -0.0028842222
      ],
      "index": 0
    }
  ],
  "model": "MODEL_ID",
  "usage": {
    "prompt_tokens": 8,
    "total_tokens": 8
  }
}

Authorization

Request Body

Response Body

200application/json

400application/json

401application/json

429application/json

500application/json

`application/json`

`application/json`

`application/json`

`application/json`

`application/json`