Ruff with Ty or Pyrefly for type checking?

Check the maturity and compatibility you need before choosing. The discussion describes Ty as an early, fast-moving tool whose behavior and diagnostics may change between releases. That can be fine for experiments or a team willing to update its setup, but it may be a poor fit if you need a stable CI check.

Pyrefly has reached a more stable release stage, which may matter more than speed for a production pipeline. Neither label settles which checker will work better on your code, so try both on a representative part of the project.

For a smaller or fairly ordinary codebase, I’d start with the simpler stack and only add complexity when a concrete gap shows up. If Ty catches the kinds of mistakes your team cares about and fits your editor and CI workflow, that may be enough.

One caveat raised in the discussion is support for Pydantic. If your project relies heavily on it, check how each tool handles your models before committing. A limitation there could matter more than general speed.

If you want the more established option, consider mypy or Pyright-based tooling too. People disagree about mypy: some find its performance and diagnostics frustrating, while others are comfortable with its behavior and prefer not to change a working setup.

So I wouldn’t switch just because a newer checker is getting attention. Compare the errors on your own code, including any framework-heavy areas, and weigh that against the stability and maintenance expectations of your team.

Pyrefly may be worth a look if its behavior fits your project and you want a checker with a stable release. One discussion mentioned using different checkers for internal code and public-facing APIs, but that’s a workflow choice, not a requirement. Running multiple checkers can also mean more configuration and more diagnostics to maintain.

There isn’t a universal winner here. The tools may differ in framework support, diagnostics, and the sort of type-checking behavior they offer, so “fast” or “stable” alone won’t tell you which is better for your codebase.

Pick a small but representative sample, run each checker with the same expectations, and look at false alarms, missed problems, setup effort, and CI behavior. If your current checker already works, you’ll need a specific benefit to make switching worthwhile.