Makepad's Relax Runs Rust Without rustc and Claims an 11-Second Game Build

Makepad, the Rust UI project, published Relax on November 11, an experimental runtime that runs a Cargo package straight from source without rustc, LLVM or a linker. Its page says a large 3D game goes from clean checkout to first frame in 11 seconds on an M4 MacBook, against about 3 minutes with cargo and rustc.

Those figures come from the project's own page, measured on 2026-11-12 with rustc 1.99.1 and LLVM 23.1.1. No independent party has reported running them yet, so treat every number below as vendor-supplied.

What Relax replaces in a Rust build

Relax swaps the whole toolchain for one Rust library of about 438,000 lines with no crates.io dependencies. It compiles each function the first time it is called, then re-optimizes hot functions in the background, moving code from an interpreter to an O1 tier and then an O3 tier.

Edits to a function body patch into the running process. Changes to struct layouts, imports, trait or impl headers and macros force a restart. The repository carries an MIT licence, 286 stars and 11 open issues, and marks the project "Experimental".

The numbers, and what they cost

The speed-up is on the build side. The price is paid at runtime, where the page says optimized code runs slower than a stock cargo build --release on most of the 21 workloads it tested.

Measure (M4 MacBook, vendor-supplied)	Relax	Baseline
Large 3D game, checkout to first frame	11 s	about 180 s
Small app, first frame	3.7 s cold, 0.41 s warm	not stated
Optimized code, geomean of 21 workloads	1.12x on x64, 1.24x on arm64	rustc release, stock
Code cache for hello	120 to 190 MB	not stated

1. Relax11 seconds
2. cargo and rustc180 seconds

Source: Makepad Relax project page, accessed 2026-11-13; baseline given as about 3 minutes

The geomean is Relax time divided by rustc time, so 1.24x means slower. Against a native-CPU build the page gives 1.22x on x64 and 1.25x on arm64. On rigid-body physics on x64 Relax is faster, at 0.43x against a stock build. The page also says kernel micro-benchmark figures are provisional because of a warm-cache tier-up bug.

Where it falls short

The page lists its own gaps. Relax uses more memory than rustc, relax check is not yet faster than cargo check, and not every crate runs. It is tested on Linux x86-64 and macOS arm64, and Windows is untested. It is for development iteration, and the page says release builds should still use cargo build --release.

Building Relax itself took 53 seconds on a 16-core x86-64 Linux box and 78 seconds on a 10-core arm64 Mac.

What the Hacker News thread says

The Hacker News thread reached 122 points. A commenter named rikarendsmp, who appears to be involved with the project, wrote that it was built with Claude and Codex over a month and "definitely isn't a one shot". Another, ricardobeat, wrote that 438,000 lines with zero dependencies is "a lot of code to maintain".

Others pointed to Cranelift, the older code generator in the same space. Commenter zamadatix argued that Relax differs because it replaces both rustc and LLVM. One user, TakerofVita, reported 10-plus minute Bevy game builds and asked for Windows support. Commenter treyd put the borrow checker at 5% to 15% of build time, which bears on how much a runtime that checks lazily can save.

The README says Makepad uses Relax in its own agent workflow, with nine agent lanes and more than 300 build and test jobs an hour. The Rust 1.100 release shipped the day after Relax appeared. Readers who want a look at how Rust projects now manage dependency risk can read our explainer on GitHub Actions lockfiles.

Makepad has not published a date for a stable release or for Windows support.