Analysis of the New Wave of Python Type Checkers
Rob Hand’s comparison of emerging type checkers is the first comprehensive look at three new Rust-based tools: ty (Astral), pyrefly (Meta), and zuban (David Halter).
Note
This analysis was published in September 2025 when ty was pre-beta (known internally as Red-Knot). ty’s typing-spec conformance has improved significantly since then, from roughly 15% at the time to 86.5% (122/141 tests, ty 0.0.65, July 2026). For current conformance scores across all major type checkers, see How do Python type checkers compare?.
Key Findings
Performance drives the rewrite. All three abandon Python implementations for Rust, targeting the bottlenecks that slow existing checkers on large codebases.
Conformance scores don’t predict usefulness. At the time of writing, ty passed only 15% of the typing conformance suite’s tests but worked well for everyday Python. The suite apparently focuses on edge cases most developers never hit.
Each tool makes different tradeoffs. Astral wants gradual adoption with fewer false positives. Meta catches more issues upfront through aggressive inference. Zuban prioritizes mypy compatibility.
For a detailed breakdown of how ty, mypy, and pyright differ, see How do mypy, pyright, and ty compare?. To get started with ty, see How to try the ty type checker.