[BUG] Vertex AI Gemini models failing with "Thought signature is not valid" error #11716

Problem (one or two sentences)

All Gemini models on Vertex AI provider are failing to generate responses, returning a 400 Bad Request error specifically mentioning an invalid "Thought signature". This started occurring after the latest model updates (including Gemini 3.1 Pro).

Context (who is affected and when)

When using Vertex AI as their API provider in Roo Code. The error appears immediately when trying to send a prompt to any Gemini model.

Reproduction steps

1. Open Roo Code settings.

2. Select Vertex AI as the API Provider.

3. Choose any Gemini model (e.g., gemini-3-pro or the new gemini-3.1-pro).

4. Type any prompt in the chat and press Enter.

5. Observe the error toast/log.

Expected result

The model should start streaming the response/thought process normally.

Actual result

An error is returned: Unable to submit request because Thought signature is not valid.. Status: INVALID_ARGUMENT (400).

Variations tried (optional)

• Tried multiple Gemini models (2.5 Pro, 3 Flash/Pro, 3.1 Pro).

Verified Vertex AI credentials and project permissions are correct.

The issue seems specific to how Roo Code handles the streaming response/thought blocks with the latest Vertex AI API changes.

App Version

3.50.4

API Provider (optional)

Google Vertex AI

Model Used (optional)

Gemini 3.1 Pro (and others in the Gemini family)

Roo Code Task Links (optional)

No response

Relevant logs or errors (optional)

Gemini generate context stream error: {
  "error": {
    "message": "{\n  \"error\": {\n    \"code\": 400,\n    \"message\": \"Unable to submit request because Thought signature is not valid.. Learn more: https://cloud.google.com/vertex-ai/generative-ai/docs/model-reference/gemini\",\n    \"status\": \"INVALID_ARGUMENT\"\n  }\n}\n",
    "code": 400,
    "status": "Bad Request"
  }
}