Nethercote reports a 5.59 percent mean wall-time reduction in the Rust compiler since December 2025, with rustdoc down 37.92 percent and Clippy cut 10 to 30 percent

Sources: Nicholas Nethercote, discussion
Summary: The post, dated 2026-07-31, measures 2025-12-03 to 2026-07-29 on rustc-perf. The 5.59 percent mean is 2.90 percent once rustdoc is excluded, and the rustdoc gain comes from Noah Lev's impl-processing work, Jakub Beranek adding rustdoc benchmarks to the PGO training set for a further 2.85 percent mean, and a change to sort impls on a short text key rather than generated HTML. Clippy is not measured by rustc-perf on CI, and the reported 10 to 30 percent runtime cut comes from combining passes so that the no-op check_foo virtual dispatches most lints never implement are skipped, with branch mispredictions down 20 to 80 percent on real examples. One new trait solver benchmark moved from 27 seconds to under one second over three months, and a separate fix that avoided two thirds of the moves of a 136-byte type and shrank it to 104 bytes, below the 128-byte threshold at which LLVM emits memcpy, cut wall-time on two pathological crates by 40 percent.
Why it matters: These are measured changes already in the compiler rather than proposals, so the gains reach any project that updates its toolchain.
Follow-up: Nethercote states he has left VectorWare and is seeking paid Rust compiler work, so track whether this measurement series continues.