structured-output-with-pydantic

Structured Output with Pydantic

Pydantic models enable type-safe, validated structured data extraction from LLM outputs.

Basic Pattern

from pydantic import BaseModel, Field
from langchain_anthropic import ChatAnthropic

class MovieReview(BaseModel):
    """A structured movie review with rating and summary."""
    title: str = Field(description="Movie title")
    year: int = Field(description="Release year")
    rating: float = Field(ge=0, le=10, description="Rating 0-10")
    summary: str = Field(description="Brief review summary")

# Initialize LLM
llm = ChatAnthropic(model="claude-sonnet-4-5-20250929")

# Create structured output wrapper
# Default: method="function_calling" (more compatible, uses tool calling)
# Alternative: method="json_schema" uses Claude's native structured output (beta)
#             Provides guaranteed schema adherence for Sonnet 4.5/Opus 4.1+
llm_with_structure = llm.with_structured_output(MovieReview)

response = llm_with_structure.invoke("Review of The Matrix")
# response is a MovieReview instance
print(response.title)   # "The Matrix"
print(response.year)    # 1999
print(response.rating)  # 8.5
# For debugging: include_raw=True returns {"raw": ..., "parsed": ..., "parsing_error": ...}

PydanticOutputParser (Alternative)

from pydantic import BaseModel, Field
from langchain_core.output_parsers import PydanticOutputParser
from langchain_core.prompts import PromptTemplate
from langchain_anthropic import ChatAnthropic

# Define the Pydantic model (same as Basic Pattern, or import from your models module)
class MovieReview(BaseModel):
    """A structured movie review with rating and summary."""
    title: str = Field(description="Movie title")
    year: int = Field(description="Release year")
    rating: float = Field(ge=0, le=10, description="Rating 0-10")
    summary: str = Field(description="Brief review summary")

llm = ChatAnthropic(model="claude-sonnet-4-5-20250929")
parser = PydanticOutputParser(pydantic_object=MovieReview)

# Create prompt with format instructions
prompt = PromptTemplate(
    template="Extract movie review.\n{format_instructions}\n\nText: {text}",
    input_variables=["text"],
    partial_variables={"format_instructions": parser.get_format_instructions()}
)

# Create chain and invoke
chain = prompt | llm | parser
result = chain.invoke({"text": "The Matrix is a 1999 sci-fi masterpiece..."})