How to disable DeepSeek V4 Flash Thinking mode in OpenCode?

[pioneerrs]

Question

I am using the OpenCode Go API key in the Immersive Translate browser extension, not directly in the OpenCode client.

My main use case is translation with the DeepSeek V4 Flash model.

It seems that Thinking / reasoning mode is enabled by default. For translation tasks, I only need the final translated text, and I do not want the model to generate or return reasoning / thinking content.

How can I disable Thinking mode in this setup?

Specifically:

• Does the OpenCode Go API support passing thinking.type = disabled?
• Is there a non-thinking variant of DeepSeek V4 Flash available?
• Is prompt-level control the only option when using the API key through a third-party extension?

[github-actions]

This issue might be a duplicate of existing issues. Please check:

• [FEATURE]:Deepseek-V4 need a "disable thinking" button #24610: Directly requests a way to disable DeepSeek V4 thinking mode (feature request for a "disable thinking" option)
• Request for support of non-thinking mode parameters in local Ollama models and documentation improvements #17510: Asks about configuring non-thinking mode parameters for models in OpenCode, including how to toggle reasoning on/off via variants

If your question is answered or addressed in one of those issues, consider adding a comment there rather than opening a new issue.

[Mrqqeat]

Currently, we only have Default, Low... and Max; we need an option to disable thinking.

[martinmr]

This is a duplicate of #24610. It will be fixed once #31795 is merged, which adds a none variant that disables thinking.

[rekram1-node]

Automated: closing question issues.

GitHub Issues are for bug reports and feature requests only — questions / support are no longer tracked here.

• For help & questions: https://discord.gg/opencode
• Billing issues with Opencode Go / Zen: help@anoma.ly

If this is a bug or feature request, please open a new issue with the appropriate template.

[jerrylususu]

Same setup here (Go API key used through a third-party client, not the OpenCode client) — I dug into this and verified a workaround against the live gateway.

For direct POSTs to https://opencode.ai/zen/go/v1/chat/completions, the parameter that actually works is reasoning_effort: "none":

{
  "model": "deepseek-v4-flash",
  "messages": [...],
  "reasoning_effort": "none"
}

I verified it live: with "reasoning_effort": "none" the response has no reasoning_content (both streaming and non-streaming), and completion_tokens drops from ~31 to ~3 on a short prompt.

Hope this helps!