Satya

Up to 9× faster than Pydantic v2. Drop-in replacement. Powered by Rust.

---

The Numbers

Raw validation:   19.4M ops/sec  (8.85× faster than Pydantic v2)
Geometric mean:    1.17× faster (across all benchmarks)
model_dump():     1.60× faster
model_dump_json:  1.30× faster

Install

pip install satya

Use

from satya import Model, Field

class User(Model):
    name: str = Field(min_length=2)
    age: int = Field(ge=0, le=150)
    email: str = Field(email=True)

user = User(name="Alice", age=30, email="alice@example.com")

user.model_dump()        # dict output — 1.6× faster
user.model_dump_json()   # JSON string — 1.3× faster

Same API as Pydantic. Change the import, get the speed.

Nested Models

class Address(Model):
    city: str
    zip_code: str

class UserProfile(Model):
    name: str
    address: Address

profile = UserProfile(name="Alice", address={"city": "NYC", "zip_code": "10001"})
print(profile.address.city)  # "NYC"

Rust writes directly into nested instance __dict__s — zero intermediate dicts.

Benchmarks

Head-to-head vs Pydantic 2.12.0 on Python 3.13 (free-threaded):

Benchmark	Satya	Pydantic	Ratio
Simple model (3 fields)	2.0M ops/s	2.2M ops/s	0.94×
Constrained (3 fields)	2.0M ops/s	2.1M ops/s	0.95×
Complex (10 fields)	946K ops/s	993K ops/s	0.95×
model_dump (simple)	2.8M ops/s	1.7M ops/s	1.60×
model_dump (complex)	1.7M ops/s	1.0M ops/s	1.63×
JSON serialize (simple)	2.1M ops/s	1.7M ops/s	1.21×
JSON serialize (complex)	1.3M ops/s	1.0M ops/s	1.30×
Raw Rust validate_check	19.4M ops/s	2.2M ops/s	8.85×

Run it yourself: python tests/benchmark_vs_pydantic.py

How It's Fast

TurboValidator — a Rust validation engine that eliminates Python overhead:

• Pre-interned keys — PyString::intern() at compile time, direct dict.get_item() with no HashMap lookup
• Return-based errors — no try/except, no string parsing. Rust returns (field, message) tuples
• Zero-allocation type checks — is_instance_of without value extraction for unconstrained fields
• u64 bitmap — field tracking without Vec allocation (up to 64 fields)
• Nested dict injection — Rust writes validated fields directly into pre-created __dict__s
• apply_default — replaces copy.deepcopy with efficient shallow copy for mutable defaults
• BLAZE ordering — cheapest validations first for fail-fast on invalid data

Validation Errors

from satya import Model, Field, ModelValidationError

class User(Model):
    name: str = Field(min_length=2)
    age: int = Field(ge=0)

try:
    User(name="A", age=-1)
except ModelValidationError as e:
    print(e.errors)  # Lazy — only created when accessed

Constraints

from satya import Model, Field

class Product(Model):
    name: str = Field(min_length=1, max_length=100)
    price: float = Field(gt=0, le=1_000_000)
    sku: str = Field(pattern=r'^[A-Z]{2}\d{6}$')
    tags: list = Field(min_items=1, max_items=10)
    email: str = Field(email=True)
    url: str = Field(url=True)

Requirements

• Python 3.9+
• No runtime dependencies

License

Apache 2.0