n8n - 💡(How to fix) Fix DeepSeek v4-flash tool calling fails in AI Agent with 400: reasoning_content must be passed back [1 comments, 2 participants]

Bug Description

When using the n8n AI Agent node with the DeepSeek Chat Model node and model deepseek-v4-flash, tool calling fails with a 400 Bad Request error:

Bad request - please check your parameters. The reasoning_content in the thinking mode must be passed back to the API.

This happens when the AI Agent invokes any connected tool.

DeepSeek's current documentation states that deepseek-v4-flash supports both thinking and non-thinking modes, and that thinking mode is enabled by default. Their Thinking Mode documentation also states that when the model performs tool calls, the assistant message's reasoning_content must be passed back to the API in the following request so the model can continue the tool-calling reasoning loop.

It looks like n8n's AI Agent / DeepSeek Chat Model integration is not preserving or re-sending reasoning_content during tool execution.

There is already a related issue/PR for deepseek-reasoner, but this now affects deepseek-v4-flash as well because DeepSeek V4 thinking mode is enabled by default and deepseek-chat is no longer a reliable workaround for some users/accounts.

To Reproduce

1. Create an AI Agent node.
2. Connect a DeepSeek Chat Model node.
3. Select model: deepseek-v4-flash.
4. Connect at least one tool to the AI Agent.
   • Example: Calculator tool, HTTP Request Tool, or any custom tool.
5. Send a prompt that forces the model to use a tool.
   • Example: "Use the tool to calculate 187 * 43 and tell me the result."
6. The agent starts the tool-calling flow.
7. n8n returns a DeepSeek API error.

Expected behavior

The workflow fails with:

Bad request - please check your parameters. The reasoning_content in the thinking mode must be passed back to the API.

This makes DeepSeek deepseek-v4-flash unusable with AI Agent tool calling in n8n.

Expected behavior

n8n should preserve the reasoning_content field returned by DeepSeek during tool-calling turns and pass it back to the DeepSeek API in the following request, together with the assistant message containing tool_calls and the tool response messages.

Alternatively, the DeepSeek Chat Model node should expose a parameter to disable thinking mode explicitly, for example:

{
  "thinking": {
    "type": "disabled"
  }
}

### Debug Info

# Debug info

## core

- n8nVersion: 2.16.1
- platform: docker (self-hosted)
- nodeJsVersion: 24.14.1
- nodeEnv: production
- database: postgres
- executionMode: scaling (single-main)
- concurrency: -1
- license: enterprise (production)
- consumerId: 8f1f5aa1-ceaf-4014-b2d6-a1791a852bea

## storage

- success: all
- error: all
- progress: false
- manual: true
- binaryMode: database

## pruning

- enabled: true
- maxAge: 336 hours
- maxCount: 10000 executions

## client

- userAgent: mozilla/5.0 (windows nt 10.0; win64; x64) applewebkit/537.36 (khtml, like gecko) chrome/147.0.0.0 safari/537.36
- isTouchDevice: false

Generated at: 2026-05-04T04:27:03.856Z

### Operating System

Ubuntu

### n8n Version

2.16.1

### Node.js Version

24

### Database

PostgreSQL

### Execution mode

main (default)

### Hosting

self hosted