DeepSeek V4Install

Summary

Send a system prompt (if provided) and required user message to DeepSeek V4 Flash or Pro, with optional reasoning controls and generation settings. Return the first completion’s answer and metadata for each input item.

Inputs

• Model — Select DeepSeek V4 Flash (default) or DeepSeek V4 Pro.
• System Message — Optional system prompt; blank values are omitted from the request.
• User Message (required) — Prompt sent as the user message.
• Thinking Mode — Disable reasoning or enable it at low, high, or maximum effort. When enabled, temperature, top P, presence penalty, and frequency penalty are not sent.
• Return Thinking Content — When thinking mode is enabled, include the response’s reasoning_content in the output. Defaults to false.
• Advanced Options — Optional generation settings: Temperature, Max Tokens, Top P, Stop Sequences, Presence Penalty, Frequency Penalty, JSON Output Mode, and Custom Base URL. Stop Sequences are split on commas. Max Tokens defaults to 2048; reasoning tokens count toward this limit. Custom Base URL overrides the credential URL.

Output shape

One output item per input item, each containing a single completion record.

The record includes content, model, usage, and finish_reason; reasoning_content is included only when thinking is enabled and Return Thinking Content is true. If the API returns no choices, the operation errors. With Continue On Fail enabled, item errors are returned as an error field.

Links

• n8n community nodes installation guide