Responses in OpenCode chat are either very choppy or none after 'Thought'

Description

I am running OpenCode with Ollama + Qwen3.5:9b/256K token window in VS Code. What I found when I ran brainstorming sessions is that the chat responses in OpenCode are often either very choppy or nothing at all after ‘Thought’ was outputted. The Gemini suggested the reason that OpenCode combines with a massive system prompt overhead that stalls OpenCode that struggles CoT text streams. So, I am wondering what the best configuration and usage heuristics (e.g., model choice, etc) so that OpenCode can actually be decently usable for such token intensive use cases.

Plugins

No response

OpenCode version

1.18.10

Steps to reproduce

1. Install BMAD method (npx bmad-method install) (Version: 6.10.0) with OpenCode selected.
2. Install Ollama (Version: 0.32.5).
3. Install qwen3.5-9b via Ollama.
4. Increase Ollama qwen3.5-9b token window to 256k.
   • create Modelfile, with
     FROM qwen3.5:9b
PARAMETER num_ctx 256000
   • run ollama create qwen3.5-9b-256k -f ./Modelfile
5. Install OpenCode CLI (Version: 1.18.10).
6. Launch VSCode (Version: 1.131.0).
7. Open VSCode terminal, enter
   ollama launch opencode --model qwen3.5-9b-256k
8. Enter the following prompt in the OpenCode (Build mode) chat box:
   bmad-agent-analyst , I prepared the project context in docs/project_context.md for you to have a brainstorming session.
9. Observed the following interactions.

Screenshot and/or share link

No response

Operating System

Mac M5 Pro

Terminal

Windows VS Code Terminal