JSON to Pydantic
Paste a JSON sample and get Pydantic models, instantly and entirely in your browser. Optional fields, nullability and nested types are inferred from the sample — and anything the sample cannot settle is reported rather than guessed.
Type definitions will appear here
Pydantic models from a sample
Generates BaseModel classes with ConfigDict(populate_by_name=True) and a Field(alias=...) wherever the JSON key had to be renamed, so the model round-trips by alias.
Verified by importing the generated module and validating a real sample against it.
What a sample cannot tell you
Types are inferred from one sample. A field absent from your sample will be absent from the output, and a field that happens to hold only whole numbers will be typed as an integer even if it can hold decimals. Paste the widest sample you have. For a guarantee rather than an inference, start from a JSON Schema.
What Pydantic output looks like
Generated from two sample records, one of which has an extra field and a null. This is the actual output, not an illustration:
from typing import Any, List, Optional
from pydantic import BaseModel, ConfigDict, Field
class Address(BaseModel):
model_config = ConfigDict(populate_by_name=True)
city: str
postcode: Optional[Any] = None
class Root(BaseModel):
model_config = ConfigDict(populate_by_name=True)
id: int
user_name: str = Field(..., alias="userName")
is_active: bool = Field(..., alias="isActive")
score: float
tags: List[str]
address: Address
last_seen: str = Field(..., alias="lastSeen")
nickname: Optional[str] = None
The three things that differ by language
Large integers
Python integers are arbitrary precision, so large identifiers survive. Pydantic v2 validates on construction, so a wrong type fails at the boundary rather than deep in your code.
Optional against nullable
Absent fields become Optional[T] = None. Because Pydantic distinguishes "not provided" from "provided as null", the generated model reflects what the samples actually showed.
Names that are not identifiers
Field(alias=...) is emitted wherever a key had to be renamed, with populate_by_name=True, so the model accepts either name and round-trips by alias.