Plan/Build mode sometimes hangs after long reasoning and never emits the final response #37609

Description

Summary

For the past 2–3 days, I've been experiencing an issue where OpenCode sometimes hangs after the model finishes a long reasoning phase.

The model successfully completes its Thought process, but never emits the final assistant response.

There is no error message or timeout. The UI simply waits indefinitely.

If I send another user message (for example: "kết quả sao rồi", "continue", or "please answer"), the model immediately returns the response that should have been sent earlier.

Environment

• OpenCode: 1.18.3
• OS: Windows 11 (latest updates)
• Terminal:
  • Windows Terminal
  • WezTerm
• Shell:
  • PowerShell 7 (installed via winget)
• Running natively on Windows (not WSL)

Models tested

This happens with multiple providers:

• DeepSeek V4 Flash Free (OpenCode Zen)
• My own API provider (OpenAI, Anthropic)

Because it occurs with different providers, it does not appear to be provider-specific.

Reproduction

The issue is not reproducible with every prompt.

However, it happens much more frequently when the model spends a long time in the Thought phase (for example 3m+ minutes).

Typical flow:

1. Start Plan mode or Build mode.
2. Send a relatively complex prompt.
3. The model performs a long reasoning phase.
4. Thought completes successfully.
5. No final assistant response is generated.
6. The session stays idle forever.
7. Send another user message.
8. The previous response immediately appears.

Expected behavior

After the Thought phase finishes, OpenCode should immediately stream the assistant response.

Actual behavior

The Thought phase completes successfully, but the assistant never starts responding.

The session appears idle until another user message is sent, which immediately causes the pending response to be emitted.

Frequency

• Started about 2–3 days ago.
• Only seems to happen after relatively long reasoning.
• Observed in both Plan mode and Build mode.

Screenshot

In the attached screenshot, the model finished a 52-second Thought phase but never produced the response until I manually sent another message asking for the result.

Plugins

OpenAI (OAuth) GPT5.4 mini, GPT 5.5 flask, DeepSeek V4 Flash

OpenCode version

1.18.3

Steps to reproduce

Start Plan mode or Build mode.
Send a relatively complex prompt.
The model performs a long reasoning phase.
Thought completes successfully.
No final assistant response is generated.
The session stays idle forever.
Send another user message.
The previous response immediately appears.

Screenshot and/or share link

Operating System

Window 11

Terminal

Windows Terminal, Wezterm