Faster Replies: Turn Off Model Reasoning
Add a one-line marker to your Agent's system message to skip a reasoning model's chain-of-thought — lower latency and fewer tokens for short-answer agents.
Written By pvdyck
Last updated 1 day ago
Why Your Agent Might Be Slow
Some AI models are reasoning models — before they answer, they run a hidden chain-of-thought. That extra thinking step is great for hard, multi-step problems, but for a short-answer chat Agent it mostly adds latency and token cost without improving the reply.
If your Agent uses a reasoning-capable model (for example minimax/minimax-m3 on the OpenRouter Chat Model node) and replies feel slow, you can switch the thinking step off.
The Marker
Add this marker anywhere in your AI Agent node's system message:
<<indie:no-reasoning>>That's it. When the platform sees the marker it:
- Removes the marker from the message before the model reads it — the model never sees the text.
- Disables the model's reasoning step for that Agent's requests.
The result is a faster, cheaper reply for the same prompt.
What It Does and Doesn't Affect
- Only reasoning-capable models are affected. On a model that has no reasoning step, the marker does nothing.
- It's opt-in. No marker means no change — your Agent behaves exactly as a standard n8n workflow would. Nothing is altered unless you add the marker yourself.
- It changes speed, not facts. The model still answers the same question with the same instructions; it just skips the internal deliberation.
When to Use It
Learn More
Reasoning is a feature of the underlying model provider. For background on what reasoning tokens are and how providers expose them, see OpenRouter's documentation: Reasoning Tokens.
Not sure which model your Agent uses? Check the OpenRouter Chat Model node in your workflow — it shows the selected model and defaults to mistralai/mistral-small-2603, which has no reasoning step.