JetBrains announced that the PyCharm 2026.2 release line includes 263 fixes and improvements, a large portion of which focus on how the development environment understands Python code. The changes include more accurate type inference, fewer false positives, more relevant autocompletion, and improvements to imports and refactoring.
The company says these changes target details that arise in daily work, particularly when using modern types and popular frameworks, rather than introducing a single major standalone feature.
More Accurate Support for SQLAlchemy 2.0
The update addresses a set of issues related to SQLAlchemy 2.0 that previously caused false positives. PyCharm now correctly handles string references to classes inside Mapped[...], and infers the type returned by Session.get() as a model object or None, rather than interpreting it as the class type itself.
The editor also now recognizes modern property setter syntax for hybrid_property using @name.inplace.setter, and handles attributes of model classes inherited through mixins, reducing reports of unexpected arguments when creating objects.
Improved Type Inference and Control-Flow Analysis
JetBrains fixed cases in which the analysis incorrectly reported that part of the code was unreachable or lost type narrowing after passing through loops. When checking a numeric union such as int | float with isinstance, the else branch is no longer treated as unreachable, and correct type narrowing continues inside while loops.
The fixes also include preserving type information when using unpacking and starred expressions, correcting the type produced by augmented assignment such as /=, and improving handling of Self and the results of constructors that use __new__, __init__, or metaclass __call__.
For enumerations, .value and .name are now inferred as precise Literal types instead of being widened to str or int. Function parameter types can also be inferred from constraints imposed by a decorator, rather than being assigned the general Any type.
More Useful Imports and Autocompletion
Automatic imports now reuse a module that has already been imported instead of suggesting a second duplicate import. If the module is already available, PyCharm prefers to reference it by its qualified name, and automatic import support for nested classes has also been added.
String targets of unittest.mock.patch() now receive autocompletion for paths that include modules, classes, and attributes. Completion for overriding built-in functions or dunter functions also now fills in the fully annotated signature and imports the required types automatically.
What Changes in Practice?
These improvements reduce the time developers spend reviewing incorrect warnings or correcting import and completion suggestions. Inline type hints, which are now clickable to navigate to the type definition, also help users understand the result of generic types directly at the point of use.
The update adds validation for format specifications inside f-strings, detecting unsupported specifications or types that do not implement __format__. In refactoring, renaming a module now also updates its related references and import statements. The name of the Refactor | Field action has also changed to Attribute to align with Python terminology for object attributes.