Connect to OpenAI® Chat Completion API from MATLAB®
model = openAIChat
model = openAIChat(systemPrompt)
model = openAIChat(___,APIKey=key)
model = openAIChat(___,Name=Value)
Connect to the OpenAI Chat Completion API to generate text using large language models developed by OpenAI.
To connect to the OpenAI API, you need a valid API key. For information on how to obtain an API key, see https://platform.openai.com/docs/quickstart.
model = openAIChat creates an openAIChat object. Connecting to the OpenAI API requires a valid API key. Either set the environment variable OPENAI_API_KEY or specify the APIKey name-value argument.
model = openAIChat(systemPrompt) creates an openAIChat object with the specified system prompt.
model = openAIChat(___,APIKey=key) uses the specified API key.
model = openAIChat(___,Name=Value) specifies additional options using one or more name-value arguments.
character vector | string scalar
Specify the system prompt and set the SystemPrompt property. The system prompt is a natural language description that provides the framework in which a large language model generates its responses. The system prompt can include instructions about tone, communications style, language, etc.
Specifying a system prompt is not supported for ModelName="o1-mini".
Example: "You are a helpful assistant who provides answers to user queries in iambic pentameter."
character vector | string scalar
OpenAI API key to access OpenAI APIs such as ChatGPT.
To keep sensitive information out of code, instead of using the APIKey name-value argument, you can also set the environment variable OPENAI_API_KEY.
For more information on connecting to the OpenAI API, see OpenAI API.
For more information on keeping sensitive information out of code, see Keep Sensitive Information Out of Code.
openAIFunction object | array of openAIFunction objects
Information about tools available for function calling, specified as openAIFunction objects.
For an example, see Analyze Scientific Papers Using ChatGPT Function Calls.
function handle
Specify a custom streaming function to process the generated output as it is being generated, rather than having to wait for the end of the generation. For example, you can use this function to print the output as it is generated.
For an example, see Process Generated Text in Real Time by Using ChatGPT™ in Streaming Mode.
Example: @(token) fprint("%s",token)
Optionally specify these properties at construction using name-value arguments. Specify PropertyName1=PropertyValue1,...,PropertyNameN=PropertyValueN, where PropertyName is the property name and PropertyValue is the corresponding value.
"gpt-4o-mini" (default) | "gpt-4" | "o1-mini" | "gpt-3.5-turbo" | ...
Name of the OpenAI model to use for text generation.
For information about currently supported models, see the OpenAI documentation at https://platform.openai.com/docs/overview.
"https://api.openai.com/v1" (default) | string scalar | character vector
Base URL, specified as a string scalar or character vector.
Specify this option to use an OpenAI-compatible API. For more information, see OpenAI-Compatible APIs.
"auto" (default) | numeric scalar between 0 and 2
Temperature value for controlling the randomness of the output. Higher temperature increases the randomness of the output.
If the model supports Temperature and Temperature is "auto", then the software uses the default temperature of the model.
"auto" (default) | numeric scalar between 0 and 1
Top probability mass for controlling the diversity of the generated output. Higher top probability mass corresponds to higher diversity.
If TopP is "auto", then the software uses the default top probability mass of the model.
[] (default) | string array with between 0 and 4 elements
Sequences that stop generation of tokens.
Example: ["The end.","And that is all she wrote."]
"auto" (default) | numeric scalar between -2 and 2
Penalty value for using a token that has already been used at least once in the generated output. Higher values reduce the repetition of tokens. Negative values increase the repetition of tokens.
If PresencePenalty is "auto", then the software uses the default presence penalty of the model.
The presence penalty is independent of the number of incidents of a token, so long as it has been used at least once. To increase the penalty for every additional time a token is generated, use the FrequencyPenalty name-value argument.
"auto" (default) | numeric scalar between -2 and 2
Penalty value for repeatedly using the same token in the generated output. Higher values reduce the repetition of tokens. Negative values increase the repetition of tokens.
If FrequencyPenalty is "auto", then the software uses the default frequency penalty of the model.
The frequency penalty increases with every instance of a token in the generated output. To use a constant penalty for a repeated token, independent of the number of instances that token is generated, use the PresencePenalty name-value argument.
10 (default) | nonnegative numeric scalar
After construction, this property is read-only.
If the OpenAI server does not respond within the timeout, then the function throws an error.
"text" (default) | "json" | string scalar | structure array
After construction, this property is read-only.
Format of the generatedOutput output argument of the generate function. You can request unformatted output, JSON mode, or structured output.
If you set the response format to "text", then the generated output is an unformatted string.
If you set the response format to "json", then the generated output is a formatted string containing JSON encoded data.
To configure the format of the generated JSON file, describe the format using natural language and provide it to the model either in the system prompt or as a user message. The prompt or message describing the format must contain the word "json" or "JSON".
For an example, see Analyze Sentiment in Text Using ChatGPT in JSON Mode.
The JSON response format is not supported for these models:
"gpt-4""gpt-4-0613""o1-mini"
To ensure that the model follows the required format, use structured output. To do this, set ReponseFormat to:
- A string scalar containing a valid JSON Schema.
- A structure array containing an example that adheres to the required format, for example:
ResponseFormat=struct("Name","Rudolph","NoseColor",[255 0 0])
Structured output is only supported for models "gpt-4o-mini", "gpt-4o-mini-2024-07-18", "gpt-4o-2024-08-06" and later.
"auto" (default) | "low" | "medium" | "high"
Verbosity of the output generated by the GPT-5 family of models, specified as "auto", "low", "medium", or "high".
If Verbosity is set to "auto", then the software chooses the default value of the model used to generate text, for example:
| Model | Default Verbosity |
|---|---|
| GPT-5 | "medium" |
| GPT-5.1 | "medium" |
| GPT-5.2 | "medium" |
For models that do not have a verbosity parameter, this property has no effect.
"auto" (default) | string | char
Amount of internal reasoning generated by the GPT-5 family of models, specified as a string scalar of character array.
If ReasoningEffort is set to "auto", then the software chooses the default value of the model used to generate text. The supported values and default value depend on the model, for example:
| Model | Default Verbosity | Supported Values |
|---|---|---|
| GPT-5 | "medium" |
"minimal", "low", "medium", "high" |
| GPT-5.1 | "none" |
"none", "low", "medium", "high" |
| GPT-5.2 | "none" |
"none", "low", "medium", "high", "xhigh" |
For models that do not have a reasoning effort parameter, this property has no effect.
character vector | string scalar
This property is read-only.
The system prompt is a natural-language description that provides the framework in which a large language model generates its responses. The system prompt can include instructions about tone, communications style, language, etc.
To set the SystemPrompt property at construction, specify the systemPrompt input argument.
Specifying a system prompt is not supported for ModelName="o1-mini".
Example: "You are a helpful assistant who provides answers to user queries in iambic pentameter."
string array
This property is read-only.
Names of the custom functions specified in the Tools name-value argument.
generate — Generate output from large language models
First, specify the OpenAI® API key as an environment variable and save it to a file called ".env". Next, load the environment file using the loadenv function.
loadenv(".env")Connect to the OpenAI Chat Completion API.
modelName = "gpt-4o-mini";
model = openAIChat("You are a helpful assistant awaiting further instructions.",ModelName=modelName)model =
openAIChat with properties:
ModelName: "gpt-4o-mini"
Temperature: "auto"
TopP: "auto"
StopSequences: [0x0 string]
TimeOut: 10
SystemPrompt: {[1x1 struct]}
ResponseFormat: "text"
PresencePenalty: "auto"
FrequencyPenalty: "auto"
FunctionNames: []
Generate text using the generate function.
generate(model,"Why is a raven like a writing desk?",MaxNumTokens=50)ans = "The phrase "Why is a raven like a writing desk?" is famously posed by the Mad Hatter in Lewis Carroll's "Alice's Adventures in Wonderland." Initially, there was no answer to this riddle, and it was meant to be nonsens"
openAIImages | openAIFunction | generate | messageHistory | ollamaChat | azureChat
- Create Simple Chat Bot
- Process Generated Text in Real Time Using ChatGPT in Streaming Mode
- Analyze Scientific Papers Using Function Calls
- Analyze Sentiment in Text Using ChatGPT in JSON Mode
Copyright 2024-2026 The MathWorks, Inc.