Grok Reasoning

Overview

Grok Reasoning uses xAI reasoning-capable models that perform internal chain-of-thought reasoning before producing a response, controlled by a reasoning-effort setting. These models are well-suited for math, coding, logic, and analytical tasks. Returns the model response along with reasoning content and token usage.

Node Type

Action — processes input items and produces output

Input / Output

Direction	Port(s)
Input	Input
Output	Output, Error

Credentials

This tool requires xAI credentials. See the Credentials Guide for setup instructions.

Parameters

Parameter	Type	Required	Default	Description
Model	options	No	Platform default	The Grok reasoning model to use.
				Options: the Grok chat models currently available — the dropdown tracks the model catalog, so it changes as xAI’s line-up changes.
System Prompt	string	No	—	System-level instructions for the model. Supports expressions like {{ $json.persona }}.
User Message	string	Yes	—	The user message or prompt to send. Falls back to item.json.message or item.json.prompt if empty. Supports expressions.
Reasoning Effort	options	No	medium	How much reasoning effort the model should spend.
				Options: low (minimal reasoning — fastest responses, lowest cost), medium (balanced reasoning effort), high (maximum reasoning — best quality, higher cost and latency)
Options	collection	No	{}	Optional output and token settings — add only the fields you want to override.
— Max Tokens	number	No	16384	Maximum number of tokens in the model response.
— Response Field Name	string	No	response	The output field name where the model response text will be stored.
Include Input	boolean	No	false	Whether to include the original input item fields in the output.
Web Search	boolean	No	false	Enable xAI’s server-side search tools. When on, Grok may search before answering and returns cited sources in the sources output field.
Search Sources	multiOptions	No	["web"]	Which xAI search surfaces Grok may use. Select both to let Grok choose per query. (shown when Web Search is on)
				Options: web (public web via the xAI web_search tool), x (X/Twitter posts via the x_search tool)
Search From Date	string	No	—	Only consider results published on or after this date. Accepts YYYY-MM-DD or a full ISO timestamp. Leave empty for no lower bound. Supports expressions. (shown when Web Search is on)
Search To Date	string	No	—	Only consider results published on or before this date. Accepts YYYY-MM-DD or a full ISO timestamp. Leave empty for no upper bound. Supports expressions. (shown when Web Search is on)
Max Concurrency	number	No	10	Maximum number of items to process concurrently.

Output Data

One output item per input item. The answer text is written to the field named by Response Field Name (response by default). The rest of the input JSON is carried over only when Include Input is on; binary data from the input item is forwarded.

• response — the model’s final answer, under whatever name Response Field Name is set to.
• reasoning — the reasoning content the model produced before answering, when the model returns it.
• model — the model that produced the answer.
• usage — the token counts reported by xAI for the request, including reasoning tokens.
• finishReason — why generation stopped, as reported by xAI.

Reference the result downstream by expression, e.g. {{ $json.response }}.

• sources is always present: an array of cited results when Web Search is on, and an empty array when it is off. Each entry carries url and title.
• usage now carries both naming schemes. prompt_tokens / completion_tokens / total_tokens are kept as aliases so existing expressions keep working, alongside the Responses API’s own input_tokens / output_tokens / num_server_side_tools_used.
• finishReason keeps the familiar stop / length / content_filter / tool_calls vocabulary. The raw Responses values are exposed additively as status and incompleteReason.
• Frequency Penalty and Presence Penalty were removed. The Responses API has no equivalent and silently ignores them, so they were retired rather than left as controls that do nothing. Saved workflows carrying those keys keep running — the values are simply ignored.

Example Configuration

Reason over a single question with the defaults:

{
  "type": "grok_reasoning",
  "parameters": {
    "userMessage": "Explain the implications of Godel's incompleteness theorems for formal systems."
  }
}

Deep analysis with maximum reasoning effort and a custom output field:

{
  "type": "grok_reasoning",
  "parameters": {
    "systemPrompt": "You are a senior systems analyst. Reason carefully and show your work.",
    "userMessage": "Identify the top three single points of failure in this architecture: {{ $json.architecture }}",
    "reasoningEffort": "high",
    "includeInput": true,
    "maxConcurrency": 5,
    "options": {
      "maxTokens": 16384,
      "responseFieldName": "analysis"
    }
  }
}

High-throughput ticket triage with minimal reasoning. Each item comes back with a single label — Bug, Feature Request or Question:

{
  "type": "grok_reasoning",
  "parameters": {
    "systemPrompt": "Classify the input as exactly one of: Bug, Feature Request, Question. Reply with only the label.",
    "userMessage": "{{ $json.ticketBody }}",
    "reasoningEffort": "low",
    "includeInput": true,
    "maxConcurrency": 20,
    "options": {
      "maxTokens": 64,
      "responseFieldName": "category"
    }
  }
}

Error Handling

Mode	Behavior
stop	Halts workflow on first error
continue	Skips failed items, passes successful ones through
errorPort	Routes failed items to Error output port

Tips

Grok Reasoning solves complex multi-step problems using an xAI Grok model. This tool is ideal for tasks requiring structured deliberation, such as math, coding, logic, and analytical challenges, rather than direct generation. It produces the model’s response, a reasoning trace, cited sources, and token usage details.