Embeddings

OpenAI-dialect embeddings with passthrough or Gemini translate.

What it is

How it works

Passthrough: parse JSON map → set model to upstream id → POST.
Google translate:

1. input must be a string or array of strings (token-id arrays rejected).
2. One string → :embedContent; many → :batchEmbedContents.
3. Map vectors back to OpenAI object: list / data[].embedding.
4. Meter prompt tokens when Gemini reports usage.

Field matrix

Examples

# OpenAI-family
curl -sS "$GW/v1/embeddings" \
  -H "Authorization: Bearer $KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "openai/text-embedding-3-small",
    "input": "hello",
    "dimensions": 512,
    "encoding_format": "float"
  }'

# Google translate
curl -sS "$GW/v1/embeddings" \
  -H "Authorization: Bearer $KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "google/text-embedding-004",
    "input": ["a", "b"],
    "dimensions": 256,
    "task_type": "RETRIEVAL_DOCUMENT"
  }'

Async / batch jobs (Google)

Long-running jobs use platform routes, not /v1/embeddings:

• POST /v1beta/models/{model}:asyncBatchEmbedContent
• /v1beta/batches*

See Platform API proxies.

Related

• Chat field parity
• Compatibility matrix