Embeddings

Google multimodal embedding models produce vectors from text, image, audio, video, and PDF inputs. The /v1/embeddings endpoint supports provider file references.

Example

curl https://llmtr.com/v1/embeddings \
  -H "Authorization: Bearer sk_your_key" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "google/gemini-embedding-2-preview",
    "input": [
      { "type": "text", "text": "Product label text" },
      {
        "type": "input_file",
        "input_file": {
          "file_id": "file_xxx",
          "mime_type": "image/jpeg"
        }
      }
    ]
  }'

Input types

Type	Description
text	Plain text
input_file	Reference to a Files API upload
image_url	Remote image URL
input_audio	Inline base64 audio

Response

{
  "object": "list",
  "data": [
    { "object": "embedding", "index": 0, "embedding": [0.023, -0.017, ...] },
    { "object": "embedding", "index": 1, "embedding": [0.041, 0.089, ...] }
  ],
  "model": "google/gemini-embedding-2-preview",
  "usage": {
    "prompt_tokens": 128,
    "total_tokens": 128
  }
}

Pricing

Google multimodal embeddings bill by the native unit shown on the model card. Upload large media through the Files API first.

Vector size

embedding-2-preview supports 768 or 3072. Truncate with dimensions:

{
  "model": "google/gemini-embedding-2-preview",
  "input": [...],
  "dimensions": 768
}