Creates a model response for the given input.
Request body
Inserts a system (or developer) message as the first item in the model's context.
An upper bound for the number of tokens that can be generated for a response, including visible output tokens and reasoning tokens.
Controls randomness in the response generation. Range: 0 to 2. Lower values produce more deterministic outputs, higher values increase variety and creativity.
Nucleus sampling parameter that controls the cumulative probability cutoff. Range: 0 to 1. A value of 0.1 restricts sampling to tokens within the top 10% probability mass.
Custom key-value pairs for storing additional information. Maximum of 16 pairs.
Enable parallel execution of multiple tool calls.
Response storage flag. Note: Currently only supports false or null values.
Enable streaming mode to receive response data as server-sent events.
Optional identifier for tracking end-user requests. Useful for usage monitoring and compliance.
Specifies the latency tier to use for processing the request.
Context truncation strategy. Supported values: auto or disabled.
Example request
{
"temperature": 1,
"top_p": 1,
"user": "user-1234"
}Response
OK
A unique identifier for the response.
The object type, which is always response.
The status of the response generation. One of completed, failed, in_progress, or incomplete.
The Unix timestamp (in seconds) of when the response was created.
Not supported. Always null.
The model used for the response.
The maximum number of tokens configured for the response.
The system instructions used for the response.
The truncation strategy used.
Metadata attached to the response.
The sampling temperature used.
The nucleus sampling parameter used.
The user identifier.
The service tier used for processing.
Whether the model can run tool calls in parallel.
Whether the response was stored.
Whether the response was generated in the background.
The number of top log probabilities returned.
The maximum number of tool calls allowed.