The question with any beta is blunt: will it break what I've got? I installed typescript@7.1.0-beta next to 7.0.2 and ran the same small projects through both. One change broke a build. Most of the rest only parse on the new compiler, and the old one rejects them as plain syntax errors. The projects, the error codes and the timings are in examples/typescript-7-1-beta-checks, and npm test re-asserts all of it on your machine.
What state is TypeScript 7.1 in right now?
Beta only. The 7.1.0-beta tag went up on 20 October 2026, and on 18 November the npm latest tag still points at 7.0.2 while beta points at 7.1.0-beta. There's no release candidate on the registry. Microsoft's announcement says the RC arrives in about a month and the final release shortly after, without a date for either.
I'm stressing that because several roundups quote exact RC and stable days. Those dates aren't in the announcement. Plan around "late November or December" and don't book a migration around a day somebody's blog invented.
Install it as a pinned dev dependency, not a floating tag:
npm install -D typescript@7.1.0-beta --save-exact
If you came here from the 7.0 line, the TypeScript 7.0 GA post covers what changed in the native compiler itself. This one is only the delta.
What did I run, and on what?
Eight cases, each a tiny project with its own tsconfig.json, run through tsc -p on 7.0.2 and on 7.1.0-beta. I used Node 22.23.5 on an Apple M1. The script prints the diagnostic codes each compiler reports, then exits 1 if a case stops matching what I describe here. Eight of eight matched on 18 November.
What actually breaks?
One thing. Here's the repro:
type Show<A extends number, B extends A> = `${B} extends ${A}`;
type Issue<A extends number, B extends A> =
A extends B ?
never :
Show<A, B>;
On 7.0.2 it's clean. On 7.1.0-beta:
index.ts(6,17): error TS2344: Type 'B' does not satisfy the constraint 'A'.
'B' is only assignable to the non-distributed 'A', but 'A' has been distributed here.
The reason is in the message. In the branch of a distributive conditional, A is a narrower type parameter for each union member, so a constraint written against the original A stops matching. B extends A was quietly relying on the old, looser behaviour.
The fix I tested: loosen the helper.
type Show<A extends number, B extends number> = `${B} extends ${A}`;
That compiled clean on 7.1.0-beta. Microsoft's notes also suggest an extra conditional check, and track the change in issue 63708 and pull request 64237. I'd call the fix cheap and the search expensive. You can't grep for it; the pattern hides in library-style generic helpers, and the first sign is a red CI run.
The one thing 7.1 breaks is a constraint that only compiled because the old branch was too loose.
Microsoft lists generated DOM lib changes as the second risk. A four-file project can't show that one, so I'm not claiming I saw it.
Does es2026 give you anything today?
Types, yes. Runtime, mostly no. --target es2026 and --lib es2026 exist on 7.1.0-beta and are rejected by 7.0.2, whose accepted target list ends at es2025 and esnext. The new lib adds Math.sumPrecise(), Error.isError(), Map and WeakMap getOrInsert(), Iterator.concat(), Uint8Array base64 and hex methods, and JSON.rawJSON().
If you stay on an es2025 lib and call one of them, 7.1 says so and points at the fix: Property 'sumPrecise' does not exist on type 'Math'. Do you need to change your target library? Try changing the 'lib' compiler option to 'es2026' or later.
Now the part the types don't tell you. I asked three Node versions what they actually ship:
| Method | Node 22.23.5 | Node 24.21.0 | Node 26.10.0 |
|---|---|---|---|
Error.isError | missing | function | function |
Map#getOrInsert | missing | missing | function |
Uint8Array#toBase64 | missing | missing | function |
Iterator.concat | missing | missing | function |
Math.sumPrecise | missing | missing | missing |
Math.sumPrecise is undefined even on 26.10.0. So the compiler will happily accept a call that throws a TypeError at startup. That's the trap with an es2026 lib: it's a promise about the language, not about your deploy target. Pair it with a runtime check or a polyfill, and keep your lib matched to the oldest runtime you ship to. I wrote about the Node side of that gap in the native TypeScript on Node post.
What about import attributes and source imports?
Both parse on 7.1.0-beta and fail on 7.0.2 with TS1005, which is a syntax error, not a type error. That's worth knowing before you share a config: a teammate on 7.0.2 won't get a helpful message.
Ambient modules can now key on the attribute instead of the file extension:
declare module "*" with { type: "css" } {
const content: CSSStyleSheet;
export default content;
}
declare module "*" with { type: "text" } {
const content: string;
export default content;
}
My test imported ./styles.css and ./notes.txt with matching attributes and both resolved without files being checked on disk. Microsoft says libraries shouldn't publish these declarations; apps declare them locally.
Source phase imports are the other one: import source wasmModule from "./add.wasm" and import.source(). They need --module set to esnext, nodenext or preserve, plus esnext in --lib or --target. I assigned the result to WebAssembly.Module and it checked clean. The compiler only types it, though; your bundler or runtime still has to implement the phase.
Are content mappers ready for Vue, Astro and Svelte?
Vue and Astro, as pre-releases. Svelte, not yet. A content mapper is a plugin that gives tsc a virtual TypeScript view of a .vue or .svelte file. You declare it in tsconfig.json:
{
"contentMappers": [
{ "package": "@vue/content-mapper", "extensions": [".vue"], "options": { "target": 3.5 } }
]
}
First the plumbing, with no mapper installed. On 7.0.2 the unknown key is ignored and the project is clean, with no warning. On 7.1.0-beta without a flag it fails with TS18068: content mappers require --runExternalCode. With the flag it fails again, TS18075, because the package isn't there to load. Microsoft's announcement still says @vue/content-mapper isn't on npm. That's out of date. It has been published since 23 October.
So on 20 November I installed it for real, in a scratch project: typescript@7.1.0-beta, @vue/content-mapper@4.0.0-alpha.0 (the npm latest tag, peer range typescript >=7.1.0-beta) and Vue 3.5. One .vue file carried two deliberate mistakes, const count: number = "not a number" in the <script setup> block and a typo, {{ label.toFixd() }}, in the template. tsc -p . --runExternalCode reported both against the right lines of the .vue file:
src/App.vue(2,7): error TS2322: Type 'string' is not assignable to type 'number'.
src/App.vue(7,15): error TS2339: Property 'toFixd' does not exist on type '"hi"'.
The same tsconfig on 7.0.2 printed nothing and exited 0. Add the flag there and 7.0.2 stops with TS5023, unknown option. I got identical output from 4.0.0-alpha.3, the alpha tag from 1 November.
Astro went the same way. @astrojs/ts-content-mapper@0.2.0 with astro@7.4.13 installed: a clean component exited 0, and const title: number = "not a number" in the frontmatter gave TS2322 at line 2, column 7. Skip installing astro and you get a wall of JSX errors (TS2875, TS7026, TS2307) from the generated code, so don't read those as the mapper failing.
Svelte isn't usable yet. Its mapper merged into the language-tools repo on 4 November, but @sveltejs/ts-content-mapper returns 404 on npm and the release pull request is still open. svelte-check 4.7.8, the current latest, still lists TypeScript ^5.0.0 || ^6.0.0 as its peer range, and I found a string in its published bundle that rewrites TS18068 to "Content mappers are not yet supported". There's also a community mdx-content-mapper 0.0.4 on npm. I didn't run it.
Read that as one broken file per framework, not a real app. I didn't test editor behaviour, build times or anything past those errors, and an alpha can change its options between releases. These two runs live in scratch projects, not in the example repo. I'd keep vue-tsc and svelte-check as your gate until a mapper reaches a stable version.
The silent pass on 7.0.2 is the part I'd watch. Someone adds contentMappers on a 7.1 branch, a CI job pinned to 7.0.2 reads a clean result, and nobody notices the mapper never ran. The flag exists because mappers are arbitrary executables, and VS Code additionally wants workspace trust.
Is the new API usable?
Yes, and the entry points are separate imports: typescript/sync, typescript/async, typescript/fs and the typescript/ast family. I wrote a probe against typescript/sync that opened my broken distributive project, took a snapshot and asked the program for semantic diagnostics. It returned one project and one diagnostic, TS2344.
One snag on the way. The release notes show using, and Node 22.23.5 throws a SyntaxError on it. Node 24.21.0 and 26.10.0 run it fine. On 22 you call snapshot.dispose() and api.close() by hand, which is what my script does. Objects are single use after disposal, so a long-lived tool should keep one API instance and create fresh snapshots.
Is it faster than 7.0?
Microsoft's numbers, --noEmit, four checker threads, against 7.0:
| Repo | 7.0 | 7.1 Beta | Speedup |
|---|---|---|---|
| bluesky | 4.48 s | 3.89 s | 1.2x |
| mui-docs | 22.2 s | 11.9 s | 1.9x |
| sentry | 19.6 s | 15.1 s | 1.3x |
| t3code-server | 26.5 s | 19.1 s | 1.4x |
| vscode | 21.6 s | 17.7 s | 1.2x |
I didn't have those repos on hand. What I could do was reuse the generated project from my TypeScript 7 benchmark post, type-check it five times per compiler and take the median, twice. At 100 modules I got 1.05x to 1.06x. At 400, 1.05x to 1.13x. At 1000, 171-173 ms against 145-148 ms, which is 1.15x to 1.19x. So the gain is real and small on code with no dependency mountain, and it grows with project size.
Memory is where the beta looks best, with mui-docs dropping from 8,093 MiB to 4,079 MiB. If your CI box keeps getting OOM-killed on a big monorepo, that's the number to test first.
Should you move to it now?
Try it on a branch, don't ship it. A beta compiler in your main pipeline buys you a 1.2x type-check you probably can't feel, and costs a tool chain that hasn't caught up. typescript-eslint's peer range, for one, still caps TypeScript below 6.1.0, so even 7.0.2 needs a workaround today; the peer range post has the ERESOLVE details and the tracking issue. Bundler support is the other thing to check before you flip anything, and the bundler guide lists where each one stands.
Here's what I'd do. Install the beta in a throwaway branch, run tsc --noEmit and read only the TS2344 lines. If there are none, you've learned your type helpers are safe and you can wait for the RC with a clear conscience. If there are some, you've got a month to fix them on your own schedule instead of during the upgrade.
I'll re-run npm test against the RC when it lands and update this post with whatever moved.