Qwen models from OpenCode Go subscription stop streaming after the first Agent step when used via GitHub Copilot BYOK

- Copilot Chat Extension Version:
- VS Code Version:
- OS Version: Windows 11 24H2
- Feature: Agent mode
- Model provider: OpenCode Go (Subscription)
- Selected models: Multiple Qwen models available in the OpenCode Go subscription
- Connection: GitHub Copilot BYOK (Custom Language Models)
- Logs: No visible errors

## What happened

I'm using GitHub Copilot with a BYOK (Custom Language Models) configuration connected to my OpenCode Go subscription.

When using any of the Qwen models included in the OpenCode Go subscription, Agent mode frequently stops immediately after the first completed step.

The response is truncated in the middle of a sentence without any error or timeout.

Example:

Finished with 1 step

Based on comparing the code between IV460...

The generation stops there and never continues.

## Comparison

Using the exact same environment, prompt, repository and BYOK configuration:

- ✅ Claude models work normally.
- ✅ GPT models work normally.
- ✅ Gemini models work normally.
- ❌ All tested Qwen models from the OpenCode Go subscription stop after the first Agent step.

This behavior appears to be reproducible across multiple Qwen models.

## Steps to reproduce

1. Configure GitHub Copilot Custom Language Models (BYOK).
2. Connect it to an OpenCode Go subscription.
3. Select any Qwen model provided by the OpenCode Go subscription.
4. Open Copilot Chat in Agent mode.
5. Ask the model to analyze or modify a project.
6. Wait for the first Agent step to complete.

## Expected behavior

The model should continue streaming until the response is complete.

## Actual behavior

The stream ends after the first Agent step.

The response is incomplete and no error message is displayed.

## Additional information

I'm not sure whether this issue originates from:

- GitHub Copilot BYOK integration,
- the OpenCode Go provider,
- or the interaction between GitHub Copilot Agent mode and the OpenCode Go implementation of Qwen models.

I'm happy to provide logs, a screen recording, or additional debugging information if needed.