Gemini Thinking

Overview

Gemini Thinking runs a Gemini model in thinking mode, so the model deliberates before it answers and the reasoning trace can be kept alongside the final response. It suits math, logic, code analysis, multi-step planning and any task where transparent deliberation improves accuracy. The thinking budget is configurable: set it to 0 to turn thinking off entirely, or -1 to let the model decide how much to spend. Optional file references can be sent as multimodal context alongside the message.

Node Type

Action — processes input items and produces output

Input / Output

Direction	Port(s)
Input	Input
Output	Output, Error

Credentials

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

Parameters

Parameter	Type	Required	Default	Description
Model	options	No	(current default)	The Gemini model to use for thinking. Pick a specific version — labels are pinned, not aliased to “latest”.
				Options: the model choices available to your workspace — pick one from the dropdown.
System Prompt	string	No	You are a helpful assistant.	System instruction that sets the behavior and persona of the model. Defaults to a generic helpful-assistant prompt — override per workflow as needed.
User Message	string	Yes	—	The message or question for Gemini to think about and answer. Supports expressions like {{ $json.prompt }}.
File URIs	string	No	—	Optional. fileUri value(s) from the Gemini File Upload tool. Pass a single URI or a JSON array for multiple files. Files are sent as multimodal context alongside the message. Falls back to item.fileUris if empty. Gemini File API retains uploads for 48 hours.
Thinking Budget	number	No	8192	Token budget for the thinking phase. Set 0 to disable thinking, -1 for dynamic budget (model decides).
Options	collection	No	{}
— Max Output Tokens	number	No	16384	Maximum number of output tokens for the response.
— Temperature	number	No	1	Sampling temperature (0-2). Lower values are more deterministic.
— Response Field Name	string	No	response	The key name under which the response text will be stored in the output item.
— Include Thinking	boolean	No	true	Whether to include the thinking trace text in the output item.
Include Input	boolean	No	false	Whether to merge the input item’s JSON fields into the output item.
Web Search	boolean	No	false	Enable Google Search grounding. When on, Gemini may search the web before answering and returns grounded sources in the “sources” output field. Works alongside File URIs and thinking.
Max Concurrency	number	No	10	Maximum number of items to process concurrently.

Output Data

One output item per input item. Binary data arriving from upstream is forwarded untouched.

The output item’s JSON contains only the fields below. The input item’s JSON is merged in only when Include Input is on:

Field	Description
response	The model’s final answer, with the thinking parts stripped out. Renamed by Response Field Name.
thinking	The reasoning trace, joined into one string. Present only when Include Thinking is on and the model actually returned thoughts — a Thinking Budget of 0 produces no trace.
model	The model that produced the response.
finishReason	Why generation stopped.
usage	Token usage metadata reported by the model. Thinking tokens are billed like any other.
sources	Grounded citations when Web Search is on — url, title, and snippet where Google supplies the excerpt. Always present; an empty array when search is off.
searchEntryPoint	Google’s Search Suggestions block, present only when returned. Their terms require displaying it wherever grounded results are shown.

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

Example Configuration

Ask one question with the defaults:

{
  "type": "gemini_thinking",
  "parameters": {
    "userMessage": "What are the key differences between TCP and UDP? When would you choose each?"
  }
}

Deep architectural reasoning with a large budget and the trace retained for review:

{
  "type": "gemini_thinking",
  "parameters": {
    "systemPrompt": "You are a senior distributed systems architect. Think through all failure modes and trade-offs before giving your recommendation.",
    "userMessage": "Design a fault-tolerant event sourcing system that must handle 100k events/sec with exactly-once delivery guarantees.",
    "thinkingBudget": 20000,
    "maxConcurrency": 2,
    "options": {
      "maxOutputTokens": 16384,
      "temperature": 0.4,
      "responseFieldName": "architectureRecommendation",
      "includeThinking": true
    }
  }
}

High-volume Q&A with thinking switched off and short answers:

{
  "type": "gemini_thinking",
  "parameters": {
    "userMessage": "{{ $json.faqQuestion }}",
    "thinkingBudget": 0,
    "includeInput": true,
    "maxConcurrency": 20,
    "options": {
      "maxOutputTokens": 256,
      "temperature": 0.5,
      "responseFieldName": "answer",
      "includeThinking": false
    }
  }
}

Let the model choose its own depth when item complexity varies across a batch:

{
  "type": "gemini_thinking",
  "parameters": {
    "systemPrompt": "Answer concisely and accurately.",
    "userMessage": "{{ $json.prompt }}",
    "thinkingBudget": -1,
    "includeInput": true,
    "maxConcurrency": 8,
    "options": {
      "temperature": 0.5,
      "responseFieldName": "geminiAnswer",
      "includeThinking": false
    }
  }
}

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

Gemini Thinking solves complex problems using a visible chain-of-thought process. It is the right tool when math, logic, code analysis, or multi-step reasoning tasks require transparent deliberation to improve accuracy. It produces sequenced thought parts followed by final response parts.