Custom provider, calling Gemini model with Google SDK results in unexpected interruptions.

[Sallyn0225]

Description

When using the Gemini model, it sometimes abruptly stops during tool calls, and at other times, it stops while thinking. However, with the same upstream source, if connected to chat tools like Chatbox or platforms like Kilo Code, it functions normally, so upstream issues can likely be ruled out.

I also used an AI gateway like AxonHub to check the logs. I can see a message indicating a stop with input but no output, and a specific response body is as follows:

{
  "candidates": [
    {
      "index": 0,
      "content": {
        "role": "model"
      },
      "finishReason": "STOP"
    }
  ],
  "responseId": "18CRacGiPIuxjuMPw_mKkAk",
  "modelVersion": "gemini-3-flash-preview",
  "usageMetadata": {
    "totalTokenCount": 22066,
    "promptTokenCount": 22066
  }
}

The "finishReason": "STOP" in the middle caught my attention, and it seems to be the cause of the abrupt termination of output.

Plugins

oh-my-opencode, opencode-axonhub-tracing

OpenCode version

1.2.4

Steps to reproduce

1. Use custom channels, use Google's SDK
2. When calling Gemini series models, there is a probability of sudden interruption. However, it will not suddenly interrupt when the model outputs text.

Screenshot and/or share link

Operating System

Windows 11 WSL2 Debian

Terminal

Windows Terminal

[github-actions]

---

This issue might be a duplicate of a recently closed issue. Please check:

• Custom provider, calling Gemini model with Google SDK results in unexpected interruptions. #13736: "Custom provider, calling Gemini model with Google SDK results in unexpected interruptions" - Exact duplicate issue from the same author that was closed due to non-compliance, but now your current issue Custom provider, calling Gemini model with Google SDK results in unexpected interruptions. #13780 is compliant with all requirements

Additionally, your issue relates to the underlying Gemini API behavior mentioned in:

• Session loop doesn't stop when model returns finish_reason=stop #11153: "Session loop doesn't stop when model returns finish_reason=stop" - Your issue directly mentions the same finishReason: STOP causing session loops

If this is a new attempt to resolve the issue from #13736, please mention that in your comments so maintainers can help track the investigation.

[github-actions]

To stay organized issues are automatically closed after 60 days of no activity. If the issue is still relevant please open a new one.