Why Use Pydantic?

• Powered by type hints
• Speed
• JSON Schema and documentation
• Custom validation / serialization logic
• Robust ecosystem (FastAPI, Django, SQLModel, Beanie, etc…)
• Extensive community

Basic Performance Tips

Opinionated JSON Loading Syntax

• Use model_validate_json(...)
• Not model_validate(json.loads(...))
• Exception: some model validators
• before, wrap

Bad:

Good:

TypeAdapter Tips

Specificity for Sequence and Mapping Type Hints

• Use list, tuple, etc instead of Sequence
• Avoid isinstance(x, Sequence)checks
• Use dict instead of Mapping
• Maximal specificity → better performance
• (in these cases)

Using Tagged Unions

Agenda

• What are tagged unions?
• Basic example
• Why use tagged unions?
• Validation performance
• Improved error handling
• JSON schema compatibility
• Advanced examples
• Callable discriminators
• Nested models and tagged unions

What are Tagged Unions?

• Special type of Union
• Also called discriminated unions
• Specify member of Union to validate against
• Based on a “tag”, also called “discriminator”
• Discriminators can be of str or Callable type

A Basic Example

A Basic Example, cont.

Efficient Validation

• Avoid validating against all union members
• Beneficial with nested models, in particular
• Crucial when models have tons of fields

​

Improved Error Handling

Improved Error Handling

Improved Error Handling

JSON Schema Compatibility

Callable Discriminators

Nested Models and Tagged Unions

Nested Models and Tagged Unions

Performance Improvements in the Pipeline

• SIMD in jiter
• Parsing JSON as validation happens
• FastModel
• Schema building speedup

​

Pydantic Logfire is in Open Beta!

• Easy to use observability platform
• “Find the needle in the stacktrace” philosophy
• Python-centric insights
• Event loop telemetry
• Profiling code and database queries
• Structured data and SQL access
• Opinionated OpenTelemetry wrapper
• Seamless Pydantic integration

​