Pydantic have very bad performance

jneeee — Sun, 13 Jul 2025 06:54:25 +0000

I have compare it with dataclasses, Nametuple, msgspec.Struct

Pydantic is the heavest package but with badest performace. I do not understand why still so many people use it.

Here is my test code:

import time
import sys
from typing import NamedTuple
from dataclasses import dataclass

from msgspec import Struct
from pydantic import BaseModel

class PointNamedTuple(NamedTuple):
    x: int
    y: int

@dataclass
class PointDataClass:
    x: int
    y: int

@dataclass(slots=True)
class PointDataClassSlots:
    x: int
    y: int


class Pointbymsgspec(Struct):
    x: int
    y: int


class Pointbypydantic(BaseModel):
    x: int
    y: int


NUM_OBJECTS = 1_000_000


def test_create_instance(func, num_objects=NUM_OBJECTS):
    start = time.perf_counter()
    _list = [func(x=i, y=i) for i in range(num_objects)]
    _size = sys.getsizeof(_list[0])
    print(f"{func.__name__}: {time.perf_counter() - start:.5f}s, {_size/1024} Bit")
    return time.perf_counter() - start, _size

print(f"Create {NUM_OBJECTS} instance, funcs: time(s): size(Bit)")
test_create_instance(PointNamedTuple)
test_create_instance(PointDataClass)
test_create_instance(PointDataClassSlots)
test_create_instance(Pointbymsgspec)
test_create_instance(Pointbypydantic)

The result:

Create 1000000 instance, funcs: time(s): size(Bit)
PointNamedTuple: 1.08123s, 0.0546875 Bit
PointDataClass: 1.09183s, 0.046875 Bit
PointDataClassSlots: 1.02618s, 0.046875 Bit
Pointbymsgspec: 0.15777s, 0.046875 Bit
Pointbypydantic: 3.67236s, 0.0703125 Bit

Dataclasses have better performace than Namedtuple

jneeee — Fri, 04 Apr 2025 11:55:18 +0000

Dataclasses exhibit superior performance compared to Namedtuple, contrary to my initial belief.

Below is the code snippet. I will create 1,000,000 instances for each class, comparing the performance of dataclasses with slot=True versus slot=False to determine which configuration is superior.

import time
import sys
from typing import NamedTuple
from dataclasses import dataclass

class PointNamedTuple(NamedTuple):
    x: int
    y: int

@dataclass
class PointDataClass:
    x: int
    y: int

@dataclass(slots=True)
class PointDataClassSlots:
    x: int
    y: int

NUM_OBJECTS = 1_000_000


def test_create_instance(func, num_objects):
    start = time.perf_counter()
    _list = [func(i, i) for i in range(num_objects)]
    _size = sys.getsizeof(_list[0])
    print(f"{func.__name__} instance {num_objects} objects: {time.perf_counter() - start:.5f} seconds, {_size} bytes")
    return time.perf_counter() - start, _size

test_create_instance(PointNamedTuple, NUM_OBJECTS)
test_create_instance(PointDataClass, NUM_OBJECTS)
test_create_instance(PointDataClassSlots, NUM_OBJECTS)

The result

PointDataClassSlots instance 1000000 objects: 0.68315 seconds, 8448728 bytes
PointDataClass instance 1000000 objects: 0.98761 seconds, 8448728 bytes
PointNamedTuple instance 1000000 objects: 1.01048 seconds, 8448728 bytes

DEV Community: jneeee

Pydantic have very bad performance

Dataclasses have better performace than Namedtuple