Limit on CandidateCount - intentional or temporarily?

Gemini API
1

Hi, quick question whether the current limit of 1 candidate per response is intentional or a temporary restriction in Google AI?

The property should allow a range 1…8 according to the documentation but anything >1 returns an HTTP 400 Bad Request.
Is that a feature of Vertex AI only?

Any comments please. Thanks.

2

Yes, all models right now support up to one candidate per response.
The documentation requires an update to reflect this - we’re tracking this and it should be fixed shortly. Thanks for flagging!

1 Like
3

Thanks for the clarification.

4

Is this limitation planned to be lifted in the near future? The docs still mention the candidateCount config as an integer between 1 and 8 (see for ex. Method: models.generateMessage  |  Google AI for Developers  |  Google for Developers) .