Run five TypeScript checkers on a fresh clone of date-fns and you get the same 309 errors. Run bun check and you get none. Same commit, same package.
Six new checkers appeared in 14 days. Their own benchmarks report up to 11.4× over TypeScript 7. The best independent number is 5.8× on VS Code. So here's a smaller test that looks at diagnostics first.
Independent numbers so far
Against TypeScript 7.0.2, at default threads:
| Checker | type-fest | Outline | VS Code | NestJS app |
|---|---|---|---|---|
| tsrs | 4.3× | 5.6× | 5.8× | 3.64× |
| bun check | 5.7× | 3.7× | 3.3× | 1.79× |
| tsc-ts | 4.5× | 3.7× | 2.2× | not run |
| tsc-rs | 3.3× | 2.5× | 2.4× | 2.46× |
The first three columns are Daniel Fry's run, on an Apple M5, Oct 19. The last is AIHUB BE's run, on a 4-vCPU GitHub runner, Oct 21. In Fry's VS Code run, tsrs falls from 5.8× to 4.2× at four threads and 3.2× at one. It also reports 10 extra errors against 7.0.2, because it tracks 7.1-dev.
Setup
Retest date: Nov 3, 2026. Machine: Apple M1 Pro, 10 CPUs, 16 GiB, macOS 27.0.1, AC power, Node v26.13.0. It wasn't idle: 1-minute load ranged from 6.5 to 28.7. ts-r, tsc-haxe and tsz weren't tested.
{ "name": "checkers", "private": true, "dependencies": { "ts702": "npm:typescript@7.0.2", "ts71beta": "npm:typescript@7.1.0-beta", "tsc-rs": "0.8.0", "tsc-ts": "0.1.0", "tsrs": "0.20.1-ts7.1.0-beta", "bun": "1.4.4" } } npm install --ignore-scripts --no-audit --no-fund (cd node_modules/bun && node install.js) export C=$PWD/node_modules 309 errors, or none
git clone https://github.com/date-fns/date-fns.git && cd date-fns git checkout a9739fb3e538 pnpm install --ignore-scripts --frozen-lockfile $C/ts702/bin/tsc -p pkgs/core --noEmit --pretty false; echo "exit $?" $C/.bin/bun check -p pkgs/core --no-script --no-pretty; echo "exit $?" pkgs/core/scripts/build/fp.ts(10,43): error TS6305: Output file '<repo>/pkgs/docs/.ts/src/index.d.ts' has not been built from source file '<repo>/pkgs/docs/src/index.ts'. ... 309 diagnostics in total ... exit 1 ✓ No type errors in 1,695 files [394.05ms] exit 0 | Checker | Errors | Exit |
|---|---|---|
| TypeScript 7.0.2 | 309 | 1 |
| TypeScript 7.1 beta | 309 | 2 |
| tsc-rs 0.8.0 | 309 | 2 |
| tsc-ts 0.1.0 | 309 | 2 |
| tsrs 0.20.1 | 309 | 2 |
| bun check 1.4.4 | 0 | 0 |
The five tsc-style outputs hash the same once the clone path is removed. Of the 309, 307 are TS6305. bun check --help says references are followed "and nothing has to be built first". After tsc -b with 7.0.2, all six exit 0. bun check counts 1,695 files and 7.0.2 counts 2,287, and this run didn't reconcile them.
The root is worse. tsc -p . --extendedDiagnostics prints Files: 0, because the root config has "include": [] and only references. All five exit 0.
Seeded errors
test-d/seeded-error.ts in type-fest at b550b410e13f:
const count: number = 'three' export const label: string = count export const greet = (name: string) => name.toUpperCase() greet(42) export const missing = greet.nope All six report the same four lines: TS2322, TS2322, TS2345 and TS2339. The exit code splits. 7.0.2 and bun check exit 1. The 7.1 beta, tsc-rs, tsc-ts and tsrs exit 2. AIHUB BE saw the same split. A CI rule on one exact code won't survive a switch.
Timings
Through /usr/bin/time -l, rotating order, 1 warm-up, median of 7. One thread is --singleThreaded, or --threads=1 for bun. date-fns ran as -p pkgs/core, after tsc -b.
| Checker | type-fest, one thread | date-fns, one thread | date-fns, default |
|---|---|---|---|
| TypeScript 7.0.2 | 25.10 s (1.0×) | 1.19 s (1.0×) | 0.72 s (1.0×) |
| TypeScript 7.1 beta | 14.28 s (1.8×) | 1.01 s (1.2×) | 0.65 s (1.1×) |
| tsc-rs | 6.96 s (3.6×) | 0.71 s (1.7×) | 0.40 s (1.8×) |
| tsc-ts | 8.33 s (3.0×) | 0.84 s (1.4×) | 0.40 s (1.8×) |
| tsrs | 7.90 s (3.2×) | 0.32 s (3.7×) | 0.20 s (3.7×) |
| bun check | 5.28 s (4.8×) | 0.70 s (1.7×) | 0.27 s (2.7×) |
Load during those columns ranged from 6.5 to 14.3. These are rough numbers from one busy machine.
Don't cite type-fest at default threads from this run. bun check read 12.1× with load between 20.4 and 28.7. A rerun stopped after four samples and read 22.9×, with 7.0.2 between 17.55 s and 83.79 s. The cause wasn't isolated. Fry's 5.7× on an M5 is the better number.
Three checks for your own repo
- Read
Files:from--extendedDiagnostics. A pass on zero files is not a pass. - Seed errors. Compare the lines and the exit code with your current checker.
- Time the whole step on an idle machine, at default threads and at one thread. A checker can't speed up a code generator it doesn't run.
Pin the version you tested. In two weeks, tsc-ts went from reporting 7.0.2 to reporting 7.1.0-beta.
The claimed numbers for every project, with their conditions, are in the maintained article.

