OpenAI Chat Model

Use OpenAI's GPT models as the LLM in your AI agent workflows

Written By pvdyck

Last updated About 5 hours ago

Connects to OpenAI's API to provide GPT model capabilities for Agent and Chain LLM workflows.

Authentication

Requires an OpenAI API key configured via Secure Vault. Token usage counts toward your OpenAI API billing.

Parameter	Description
Model	The GPT model to use (see table below).
Temperature	Controls randomness (0 = deterministic, 1 = creative). Default: 0.7.
Max Tokens	Maximum number of tokens in the response.
Top P	Nucleus sampling (0-1). Alternative to temperature.
Frequency Penalty	Reduces repetition of frequent tokens (-2.0 to 2.0).
Presence Penalty	Encourages topic diversity (-2.0 to 2.0).

Model	Best For
GPT-4o	Most capable, multimodal (text + vision)
GPT-4o Mini	Fast, affordable, good for most tasks
GPT-4 Turbo	High capability with large context
GPT-4	Strong reasoning, older model
GPT-3.5 Turbo	Cheapest, basic tasks only

Model availability depends on your OpenAI API plan.

Option	Description
JSON Mode	Forces the model to output valid JSON. Set `response_format: json_object`.
Seed	For reproducible outputs. Same seed + input = same output (beta).
System Prompt	Persistent instructions for the model's behavior.
Timeout	Maximum request duration in milliseconds. Prevents hanging on slow responses.
Max Retries	Number of automatic retry attempts if the API request fails.

Parent Node	Description
Agent	Powers the agent's reasoning and tool-calling decisions.
Chain LLM	Generates text responses from prompts.
Information Extractor	Extracts structured data from text.
Text Classifier	Classifies text into categories.
Sentiment Analysis	Analyzes text sentiment.

Must be connected to a parent node — cannot run standalone.
No streaming support in the Worker environment.
JSON Mode requires the word "JSON" somewhere in your prompt for reliable output.

Use GPT-4o Mini for most workflows — fast and cost-effective.
Use GPT-4o for complex reasoning, vision tasks, or when accuracy is critical.
Enable JSON Mode when combining with Output Parser Structured for reliable structured output.
Set Frequency Penalty to 0.5-1.0 to reduce repetitive responses.
Sub-node expression note: As a sub-node, expressions like {{ $json.name }} always resolve to the first input item, not each item individually. Use the parent node's batch processing if you need per-item model calls.
Consider using OpenRouter Chat Model instead for access to GPT models plus other providers through a single credential.