Docs
Embeddings
POST /v1/embeddings, OpenAI format.
Distributed processing. Jobs run on a network of vetted consumer GPUs and Macs owned by independent contributors. Do not send personal or sensitive data. Read the data policy.
{ "model": "bge-m3", "input": ["first document", "second document"], "encoding_format": "float" }
| Field | Notes |
|---|---|
model | bge-m3 (multilingual, 1024-d, 8,192 tokens) or bge-small-en-v1.5 (English, 384-d, 512 tokens) |
input | A string or an array of up to 2,048 strings. Token arrays are refused (unsupported_parameter), because we count tokens ourselves with the model's tokenizer. |
encoding_format | float (default) or base64 |
dimensions | Optional truncation where the model supports it |
usage.prompt_tokens in the response is the count from our metering service. That is what you're billed for.
Tips
- Batch inputs: one request with 500 short strings is far cheaper to run than 500 requests.
- Embeddings are compared against a reference by cosine similarity during verification, so tiny numeric differences between devices are expected and harmless.