Structured outputs

Ask the model to return JSON validated against a Pydantic schema.

1. Define the schema

   from pydantic import BaseModel
   
   class Weather(BaseModel):
       location: str
       temperature_c: float
       condition: str
   

2. Call generate_object

   weather.py

   from ai_sdk import openai, generate_object
   
   model = openai("gpt-4.1-mini", temperature=0)
   
   res = generate_object(
   model=model,
   schema=Weather,
   prompt="Return the current weather in Berlin as JSON (no extra text).",
   )
   print(res.object)
   
   

Expected output (Pydantic instance):

Weather(location='Berlin', temperature_c=18.3, condition='Sunny')

If the model sends invalid JSON, the helper retries parsing and raises a clear exception if validation ultimately fails.