API developer functions - now with an output schema describing expected tool response

This feature does not appear anywhere in documentation.

It is being offered by the AI function creation helper in the chat playground

Function definitions may include an output_schema that describes their structured result.

That is followed immediately by a “learn more” link, with nothing to learn about the schema top-level field.

output_schema

With no documentation, whether this new function content is seen by the AI or could improve the AI’s sending and knowledge of what can be fulfilled by a function is not clear. Also, which chat endpoint and which models tolerate or use data.

Some test scripting showing the increase in billing could let you know if the AI model is receiving extra descriptive tokens. Then you could evaluate the quality of calling, or the quality of answering about tools->functions (now generally resulting in a refusing AI). Gap-filling reverse-engineered technical report writing is not on my plate right now.

Example

I describe to the AI that the tool will return a JSON with some fields, that gets written to our favorite weather function example.

{
  "name": "get_weather",
  "description": "Get the weather for a city with temperature in Celsius or Fahrenheit.",
  "strict": true,
  "parameters": {
    "type": "object",
    "properties": {
      "city": {
        "type": "string",
        "description": "Name of the city to get weather for."
      },
      "units": {
        "type": "string",
        "description": "Temperature units, either 'C' for Celsius or 'F' for Fahrenheit.",
        "enum": [
          "C",
          "F"
        ]
      }
    },
    "required": [
      "city",
      "units"
    ],
    "additionalProperties": false
  },
  "output_schema": {
    "type": "object",
    "properties": {
      "location": {
        "type": "string",
        "description": "Name of the city or location."
      },
      "temperature": {
        "type": "string",
        "description": "Temperature in the specified units."
      },
      "units": {
        "type": "string",
        "description": "Temperature units, 'C' or 'F'."
      },
      "conditions": {
        "type": "string",
        "description": "Weather conditions description."
      }
    },
    "required": [
      "location",
      "temperature",
      "units",
      "conditions"
    ],
    "additionalProperties": false
  }
}

gpt-5-mini doesn’t seem to know:

The user is asking about the kind of information I can provide when retrieving weather data for a city. I think they want a list of typical data from a weather API or service. I should cover categories like current conditions, forecasts, alerts, historical data, air quality, meteorological metadata, and more. It’s important to be concise but comprehensive, and I might offer to provide an example or ask for a specific city or units. I’ll keep my formatting user-friendly!

gpt-5.4-mini does:

I need to provide a general answer about the weather tool’s capabilities. There’s no need for the tool itself, but I should mention that it likely gives temperature in either Celsius or Fahrenheit, along with conditions. The tool description suggests it returns the city’s weather, including the chosen temperature unit, but I need to be careful not to invent specifics. I’ll focus on crafting a concise answer that reflects that accurately.