ts-rust uses AI models to port TypeScript 7 compiler to Rust, says it runs about twice as fast as Go version

A developer has released ts-rust, an experimental Rust port of the TypeScript 7 compiler. The project lives under the pingdotgg account on GitHub. The author says he wanted to test whether AI models could port the TypeScript compiler, type checker and language server to Rust, and concluded that they can.

He says the cost was high. Over $420,000 in API-priced tokens went into the effort, though he adds that you could probably do it for around $20,000. He also writes that he has never read a line of the code, and he warns that he does not know if the install will work. The release is described as early.

The author says the first attempts used OpenAI models. By his account, GPT-5.6 Sol and GPT 6 Astra together used over $400,000 in API-priced tokens, wrote more than 1.3 million lines of Rust over several months, and never got past about 84% compatibility.

He then tried Claude's Opus 5.5, partly because his Claude Code limits were barely being used. He says it produced a working first version in 10 hours. He had assumed it was reusing the earlier code, but it started from scratch. He says it got further than the earlier models in a tenth of the time.

He let it keep going. Total spend was about $24,047 in API terms over two weeks. Because he used his own Claude accounts, he says this came to roughly 925% to 983% of his $200 plan's weekly limits.

The README says ts-rust is a direct port of Microsoft's native TypeScript compiler, which is written in Go. It keeps Go's algorithms and behavior, and offers the same command line (tsc), language server and API. Everything in the README below a section called "The Slop Line" was written by the author's AI tools, not by him.

The port is pinned to one upstream revision, from 29 September 2026, labeled TypeScript 7.1.0-dev. According to the README, all 181,711 ported Go tests pass. It also says TanStack Query core and Hono produce diagnostics identical to Go's, and that across 120 open-source repos the command-line output differs from Go's only in the known problems and where Go's own output varies between runs.

On speed, the README reports that type checking on 60 open-source projects takes about half of Go's time, as a geometric mean (a kind of average suited to ratios). In a table of six apps, tsc-rs is 1.61 times faster than tsc 7 overall. On VS Code, which has 3.75 million lines checked, it took 4.20 seconds against 6.84 for tsc 7 and 54.56 for tsc 6, the older JavaScript compiler. These figures come from the project's own benchmarks on an Apple M4 Pro, and the draft does not say anyone else has repeated them.

The README also compares it with bun check, a type checker in Bun. Bun is faster in most cases, about 2.95 times faster than tsc 7 on average. But it reports a few errors no other checker does, and it lacks Effect diagnostics. tsc-rs has Effect language service diagnostics built in, so Effect projects need only one pass. On the T3 Code projects, it took 11.13 seconds with those diagnostics against 21.07 for tsc 7 with the Effect patch.

The README lists known problems. In some monorepos, tsc-rs can write output for more files than tsc and report a rootDir error (TS6059). In tsc -b builds, a project that imports another's output without a project reference can read old or missing output. In the editor, memory grows slowly, about 20 MiB per 1,000 edits. And tsc-rs --version prints the TypeScript version it ports, not the npm version.

It currently supports Linux x64 and macOS arm64. Windows and Linux arm64 are not available yet. The npm package is called tsc-rs, so it does not clash with the typescript package.