The tsc-rs project is a Rust version of the TypeScript compiler that Claude wrote in about two weeks for roughly $24,000. Developer Theo Browne open-sourced it on October 8, 2026, and says it passes all 181,711 ported compiler tests.
It also claims to beat Microsoft’s own Go-based compiler on speed. Here is what is confirmed, what is vendor-reported, and whether you should touch it yet.
Here are the facts most readers want first.
- Released: October 8, 2026, as open source under the MIT license
- Built by: Theo Browne of Ping Labs, using Claude Opus 5.5
- Cost and time: about $24,047 and two weeks
- Speed claim: 1.61x faster than tsc 7 on average (author’s own benchmarks)
- Platforms: Linux x64 and macOS arm64 only
Treat it as an early release. The author’s README lists open bugs, and nobody outside the project has verified the speed numbers yet.
What tsc-rs actually is
Microsoft recently rewrote the TypeScript compiler in Go, which ships as TypeScript 7. The tsc-rs project ports that Go code to Rust. According to AICoder, it covers the compiler, the type checker, and the language server, and it is pinned to TypeScript 7.1.0-dev.
The package is on npm as tsc-rs, and the source lives in the pingdotgg/ts-rust repository on GitHub. You can try it with npm install -D tsc-rs and then run npx tsc-rs -p tsconfig.json.
The most unusual line is in the README itself. Browne writes that he has never read a line of this code. That is a bold claim for a project with more than 1.3 million lines of Rust and nearly 6,000 commits.
The $24K price tag and the failed attempts before it
The cost is the story here. AI Weekly, citing the README, reports that a Codex-based attempt earlier in the project cost more than $400,000 and stalled at about 84% compatibility. Claude then started over from scratch.
The table below compares the two runs using the figures those reports give.
| Earlier attempt | Claude Opus 5.5 run | |
|---|---|---|
| Reported cost | Over $400,000 (API-priced) | About $24,047 |
| Compatibility | About 84%, then stalled | All 181,711 ported tests pass |
| Time to first working version | Not reported | About 10 hours |
| Time to finish | Not finished | About two weeks |
One caution: AICoder says the report does not make clear whether the $24K is API pricing or a subscription figure. Compare the two numbers loosely, not to the dollar.
How fast is tsc-rs compared with tsc 7?
The author ran the benchmarks himself on an M4 Pro Mac using hyperfine. That makes them useful as a hint, not as proof. Independent tests will tell us more in the coming weeks.
These are the headline results from the README.
| Test | tsc-rs | Comparison |
|---|---|---|
| Six real-world apps (geometric mean) | 1.61x faster | Versus tsc 7 |
| Same apps (geometric mean) | 11.4x faster | Versus tsc 6 |
| T3 Code project (median of five runs) | 7.25 seconds | tsc 7: 16.10 seconds |
| VS Code, about 3.75M lines | 4.20 seconds | tsc 7: 6.84 seconds |
There is one fair counterpoint. The README says bun check is far faster, but it skips Effect diagnostics, which tsc-rs includes. So the speed gap depends on how much checking you need.
Should you use tsc-rs in your own project?
Not for production builds yet. The project is days old, and the README is upfront about its rough edges.
These are the known issues so far.
- Memory grows by about 20 MiB per 1,000 edits in long editor sessions.
- Timing varies in monorepos, and some hit a TS6059 error.
- Multi-project builds with
tsc -bcan produce stale output when project references are missing.
The stale-output bug matters most. A build that passes while serving old results is worse than a slow build.
A safer first step is to run tsc-rs next to your normal compiler in CI and compare the diagnostics. The README says output matched the Go version on TanStack Query core and Hono. If you care about AI-written code safety in general, our look at the Anthropic OSS scanner for open-source maintainers is a useful companion read.
Why a Claude-written compiler port matters beyond TypeScript
A compiler port is a hard test. The target behavior is huge, the spec is the existing code, and tiny mistakes break real builds. A 181,711-test suite gives the model a clear pass-or-fail signal, and that is likely why this worked.
The lesson is about method more than model. Big, well-tested codebases are now realistic targets for AI-driven rewrites. If you are weighing model costs for similar agent work, see our breakdown of Claude Haiku 5.5 pricing.
Still, a passing test suite is not the same as trusted code. With no human review, bugs outside the tests can hide for a long time.
Frequently Asked Questions
Is tsc-rs an official Microsoft project?
No. It is an independent open-source port by Theo Browne and Ping Labs. It is based on Microsoft’s Go version of the TypeScript compiler.
How do I install tsc-rs?
Run npm install -D tsc-rs in your project, then npx tsc-rs -p tsconfig.json. It currently supports Linux x64 and macOS arm64 only.
Is tsc-rs really faster than the Go TypeScript compiler?
The author reports 1.61x faster on average. Those results come from his own M4 Pro tests, so wait for independent benchmarks before you rely on them.
Does tsc-rs work on Windows?
Not yet. The first release ships binaries only for Linux x64 and macOS arm64.
Our verdict: test it in CI, skip it in production
The $24K figure will get the headlines, but the real value for you is a possible faster type check later this year. Try tsc-rs on a side branch, compare its output with your current compiler, and report bugs. Wait for Windows support and a few independent benchmarks before you switch anything that ships to customers.
Leave a Reply