New is_query Parameter on the Embed APIs for Optimised Retrieval Vectors
A query and a document are not the same kind of text, and embedding them identically costs you retrieval quality. The embed and batch embed endpoints now take an is_query parameter.
01 One parameter, two roles
Set is_query=1 when the text you are embedding is a search query rather than a document. You get a vector optimised for retrieval against the vectors you stored at index time. Leave it off, as before, when you are embedding content to be stored.
| Value | Use it for |
|---|---|
is_query=1 | Embedding a search query — returns an optimised retrieval vector. |
| omitted | Embedding content for indexing, exactly as before. |
The parameter is available on both the single
embed endpoint and the batch embed endpoint.