DeepSeek Reasoner Chat Model icon

DeepSeek Reasoner Chat Model

Chat model for DeepSeek R1/Reasoner with proper reasoning_content handling for tool calls. Fixes the missing reasoning_content error. Connect this to an AI Agent node as the Chat Model.

DeepSeek Reasoner Chat Model

AI-generated

Summary

Connects a DeepSeek chat model to an n8n AI Agent or chain as its language model. It sends the supplied conversation to DeepSeek, can support tool calls, and returns a LangChain chat model response with token usage when the API provides it.

Inputs

  • Model — Choose DeepSeek Chat, Reasoner, V4 Pro, or V4 Flash; defaults to DeepSeek Chat.
  • Enable Thinking — Defaults to enabled. When enabled, requests thinking mode and applies the selected reasoning effort; when disabled, sampling and penalty settings are sent instead.
  • Options → Response Format — Choose text (default) or JSON Object. For JSON Object, include the word “json” in the prompt.
  • Options → Reasoning Effort — Choose High (default) or Max; sent when thinking is enabled.
  • Options → Max Tokens — Defaults to 8192 and includes hidden reasoning tokens. Set to 0 to send the model maximum of 384000.
  • Options → Temperature, Top P, Frequency Penalty, Presence Penalty — Optional generation controls. The source sends these only when thinking is disabled; DeepSeek Reasoner ignores temperature and top_p.
  • Options → Max Retries — Defaults to 2 retries for API failures; non-rate-limit 4xx errors are not retried.

Output shape

An AI Language Model connection that generates chat responses or tool calls for a connected AI Agent or chain.

The model is stateless between generation calls and uses the conversation messages passed by LangChain. Responses include text or tool-call data, retain reasoning content in message metadata, and include token usage when returned by the API. The node requires DeepSeek API credentials.

Links

Discussion