CandidateCount > 1 returns 400 "Multiple candidates is not enabled for this model" on Gemini 3.x models (regression)

[AmitTzah]

candidateCount worked fine on Gemini 2.x models but any value above 1 now returns a 400 error on all Gemini 3.x models I've tested, including the stable GA gemini-3.1-flash-lite. The parameter is still present and undeprecated in the SDK types, and there's been no changelog entry or announcement about this change.

Environment details

• Programming language: TypeScript / JavaScript
• OS: Windows 11
• Language runtime version: Node.js v24.1.0
• Package version: @google/genai 2.4.0

Steps to reproduce

1. Run the following:

import { GoogleGenAI } from "@google/genai";

const ai = new GoogleGenAI({ apiKey: process.env.GEMINI_API_KEY });

const response = await ai.models.generateContent({
  model: "gemini-3.1-flash-lite",
  contents: "Give me a short poem about the sea.",
  config: {
    candidateCount: 2,
    temperature: 1.0,
  },
});

console.log(response.candidates);

2. Returns:

{
  "error": {
    "code": 400,
    "message": "Multiple candidates is not enabled for this model",
    "status": "INVALID_ARGUMENT"
  }
}

Same code with gemini-2.5-flash works fine. Also reproduced on gemini-3-flash-preview and gemini-3.1-pro-preview.

Same behavior also reported in python-genai: googleapis/python-genai#1723

[Varun-S10]

Hi @AmitTzah, Thank you for raising this issue and providing the reproducible code. We were able to reproduce the behavior on our end. We truly appreciate you flagging this. we are filing a bug internally.

[AmitTzah]

@Varun-S10 Any update on this? This is basic functionality that's broken on an already GA release. It makes the new models unusable for our use case since we need multiple outputs per input, and making n separate calls isn't practical due to cost and latency.