Python Dataclasses vs Pydantic: Which Should You Use?

Both give you classes with typed fields and less boilerplate. The key difference: dataclasses store data, Pydantic validates and converts it.

Dataclasses

from dataclasses import dataclass, field

@dataclass(frozen=True, slots=True)
class Money:
    amount: int        # paise
    currency: str = "INR"

@dataclass
class Cart:
    items: list[str] = field(default_factory=list)

m = Money(49900)
m == Money(49900)      # True: __eq__ generated
Money("lots")          # no error! type hints aren't checked at runtime

• Part of the standard library, no dependency.
• Very fast to create; ideal for internal data passed between your own functions.
• frozen=True for immutable values, slots=True for lower memory use.

Pydantic

from pydantic import BaseModel, Field, EmailStr

class SignupRequest(BaseModel):
    email: EmailStr
    age: int = Field(ge=13)
    newsletter: bool = False

req = SignupRequest.model_validate({"email": "a@b.com", "age": "21"})
req.age               # 21 - the string was converted to int
SignupRequest(email="nope", age=5)  # ValidationError listing both problems

req.model_dump_json() # serialise to JSON

• Validates and coerces data from untrusted sources.
• JSON parsing and serialisation built in, plus JSON Schema generation.
• Powers FastAPI request and response models.

How to choose

• Data crossing a boundary (HTTP requests, config files, environment, third-party APIs, LLM output) - Pydantic.
• Internal domain objects and value types - dataclasses.
• Pydantic can validate dataclasses too (pydantic.dataclasses.dataclass or TypeAdapter), so you can mix both.

A common pattern: Pydantic models at the edges of your app, converted into plain dataclasses for the core business logic.