- Joined
- Jan 22, 2026
- Messages
- 101
- Reaction score
- 926
The new vinext framework has reduced the size of client bundles by 56% in the test application.
Cloudflare showed an unusual experiment at the intersection of AI and development and immediately came into the spotlight of the frontend community. The company's engineer, along with the AI model, assembled vinext, a compatible alternative to Next, in a few days.Vite-based js, which can be deployed to Cloudflare Workers with a single command. The authors of the project claim a faster build and noticeably smaller client bundles compared to Next.js in the test scenario.
The appearance of vinext Cloudflare explains the old problem of Next.js in serverless environments. The framework itself is convenient for development, but when running on different platforms, engineers often have to adjust the build results to a specific infrastructure. OpenNext is already being used for this task, and Cloudflare explicitly recognizes the value of the project, but calls the approach fragile. The reason is simple: OpenNext is forced to disassemble and repackage the internal build output Next.js after changes, but local development still remains tied to Node.js . This combination makes it more difficult for developers to test Cloudflare services and platform APIs even at the dev build stage.
Instead of another adapter, the team decided to repeat the Next API surface.js on top of Vite. vinext does not use the internal assembly of Next.js does not wrap Turbopack. The project re-implements key mechanics through the Vite plugin, including routing, SSR, React Server Components, server actions, caching, and middleware. According to Cloudflare's description, the migration looks pretty soft: it's enough for the developer to install the package and replace next with vinext in the scripts, and the app and pages folders, as well as next.config.js, continue to work without rewriting.
The most notable part of the announcement is related to build performance. In a test application with 33 routes, vinext on Vite 7 and Rollup assembled the project faster than Next.js 16.1.6 with Turbopack. The version on Vite 8 and Rolldown showed an even bigger gap. Cloudflare also reported a reduction in the size of the client's gzipped bundle by about 56-57% in the same test. The team separately clarifies the boundaries of the comparison: we are talking about compilation and build time, not production server performance, and the results are obtained on the same application.
So far, vinext is primarily focused on Cloudflare Workers. For deployment, the company prepares the vinext deploy team, which builds the application, creates the Worker configuration and sends the project to production. The project supports App Router and Pages Router, client hydration and navigation. For caching, Cloudflare offers a KV-based processor with ISR out of the box, and has made the cache layer removable to connect other storages, including R2, if desired. This approach is especially interesting for teams that need Durable Objects, AI bindings, and other Cloudflare services already at the local development stage.
Cloudflare is not trying to pass vinext off as a ready-made replacement for Next.js for all scenarios. The company explicitly writes that the project is experimental, has been in existence for less than a week and has not yet been tested by large combat loads. At the same time, the team shows serious testing training: more than 1,700 tests on Vitest, hundreds of E2E checks on Playwright, and the transfer of tests from the Next ecosystem.js and OpenNext, and compatibility with the Next API.js 16 is estimated at 94Cloudflare is not trying to pass vinext off as a ready-made replacement for Next.js for all scenarios. The company explicitly writes that the project is experimental, has been in existence for less than a week and has not yet been tested by large combat loads. At the same time, the team shows serious testing training: more than 1,700 tests on Vitest, hundreds of E2E checks on Playwright, and the transfer of tests from the Next ecosystem.js and OpenNext, and compatibility with the Next API.js 16 is estimated at 94%. Cloudflare cites the site as an early example CIO.gov , where vinext is already in production and, according to the company, provides benefits in terms of assembly and bundle size.
Separately, Cloudflare offered a way to speed up the launch of large sites without completely pre-generating all pages. Currently, vinext is able to ISR after the first request, but it does not yet support the classic static pre-render at the build stage. Instead of full pre-rendering, the team is tesSeparately, Cloudflare offered a way to speed up the launch of large sites without completely pre-generating all pages. Currently, vinext is able to ISR after the first request, but it does not yet support the classic static pre-render at the build stage. Instead of full pre-rendering, the team is testing the Traffic-aware Pre-Rendering mode. During the deployment, the system analyzes Cloudflare traffic data and renders in advance only the pages that give the main stream of visits. For directories with tens or hundreds of thousands of URLs, this approach can dramatically reduce the build time, because the "hot" pages are prepared in advance, and the rest goes to the SSR with subsequent caching via the ISR.
The loudest thesis in the announcement concerns the development process. Cloudflare claims that almost all of the vinext code was generated by an AI model under the supervision of one engineer, and the cost of tokens was about $ 1,100. According to the author of the project, tThe loudest thesis in the announcement concerns the development process. Cloudflare claims that almost all of the vinext code was generated by an AI model under the supervision of one engineer, and the cost of tokens was about $ 1,100. According to the author of the project, the first commit appeared on February 13, and by the end of the day, basic SSR, middleware, server actions, and streaming for both routers had already started working. Next, the team completed compatibility, tests, and complex cases.
At the same time, Cloudflare does not describe the operation as a "magic button". The engineer set the architecture, set priorities, stopped unsuccessful decisions, and kept the process within strict quality gates. The cycle looked familiar to any developer: setting a task, generating code and tests, running checks, corrections and repetition. Part of the pull request review was performed by agent-based tools, but the final decisions were left to the individual.
Cloudflare showed an unusual experiment at the intersection of AI and development and immediately came into the spotlight of the frontend community. The company's engineer, along with the AI model, assembled vinext, a compatible alternative to Next, in a few days.Vite-based js, which can be deployed to Cloudflare Workers with a single command. The authors of the project claim a faster build and noticeably smaller client bundles compared to Next.js in the test scenario.
The appearance of vinext Cloudflare explains the old problem of Next.js in serverless environments. The framework itself is convenient for development, but when running on different platforms, engineers often have to adjust the build results to a specific infrastructure. OpenNext is already being used for this task, and Cloudflare explicitly recognizes the value of the project, but calls the approach fragile. The reason is simple: OpenNext is forced to disassemble and repackage the internal build output Next.js after changes, but local development still remains tied to Node.js . This combination makes it more difficult for developers to test Cloudflare services and platform APIs even at the dev build stage.
Instead of another adapter, the team decided to repeat the Next API surface.js on top of Vite. vinext does not use the internal assembly of Next.js does not wrap Turbopack. The project re-implements key mechanics through the Vite plugin, including routing, SSR, React Server Components, server actions, caching, and middleware. According to Cloudflare's description, the migration looks pretty soft: it's enough for the developer to install the package and replace next with vinext in the scripts, and the app and pages folders, as well as next.config.js, continue to work without rewriting.
The most notable part of the announcement is related to build performance. In a test application with 33 routes, vinext on Vite 7 and Rollup assembled the project faster than Next.js 16.1.6 with Turbopack. The version on Vite 8 and Rolldown showed an even bigger gap. Cloudflare also reported a reduction in the size of the client's gzipped bundle by about 56-57% in the same test. The team separately clarifies the boundaries of the comparison: we are talking about compilation and build time, not production server performance, and the results are obtained on the same application.
So far, vinext is primarily focused on Cloudflare Workers. For deployment, the company prepares the vinext deploy team, which builds the application, creates the Worker configuration and sends the project to production. The project supports App Router and Pages Router, client hydration and navigation. For caching, Cloudflare offers a KV-based processor with ISR out of the box, and has made the cache layer removable to connect other storages, including R2, if desired. This approach is especially interesting for teams that need Durable Objects, AI bindings, and other Cloudflare services already at the local development stage.
Cloudflare is not trying to pass vinext off as a ready-made replacement for Next.js for all scenarios. The company explicitly writes that the project is experimental, has been in existence for less than a week and has not yet been tested by large combat loads. At the same time, the team shows serious testing training: more than 1,700 tests on Vitest, hundreds of E2E checks on Playwright, and the transfer of tests from the Next ecosystem.js and OpenNext, and compatibility with the Next API.js 16 is estimated at 94Cloudflare is not trying to pass vinext off as a ready-made replacement for Next.js for all scenarios. The company explicitly writes that the project is experimental, has been in existence for less than a week and has not yet been tested by large combat loads. At the same time, the team shows serious testing training: more than 1,700 tests on Vitest, hundreds of E2E checks on Playwright, and the transfer of tests from the Next ecosystem.js and OpenNext, and compatibility with the Next API.js 16 is estimated at 94%. Cloudflare cites the site as an early example CIO.gov , where vinext is already in production and, according to the company, provides benefits in terms of assembly and bundle size.
Separately, Cloudflare offered a way to speed up the launch of large sites without completely pre-generating all pages. Currently, vinext is able to ISR after the first request, but it does not yet support the classic static pre-render at the build stage. Instead of full pre-rendering, the team is tesSeparately, Cloudflare offered a way to speed up the launch of large sites without completely pre-generating all pages. Currently, vinext is able to ISR after the first request, but it does not yet support the classic static pre-render at the build stage. Instead of full pre-rendering, the team is testing the Traffic-aware Pre-Rendering mode. During the deployment, the system analyzes Cloudflare traffic data and renders in advance only the pages that give the main stream of visits. For directories with tens or hundreds of thousands of URLs, this approach can dramatically reduce the build time, because the "hot" pages are prepared in advance, and the rest goes to the SSR with subsequent caching via the ISR.
The loudest thesis in the announcement concerns the development process. Cloudflare claims that almost all of the vinext code was generated by an AI model under the supervision of one engineer, and the cost of tokens was about $ 1,100. According to the author of the project, tThe loudest thesis in the announcement concerns the development process. Cloudflare claims that almost all of the vinext code was generated by an AI model under the supervision of one engineer, and the cost of tokens was about $ 1,100. According to the author of the project, the first commit appeared on February 13, and by the end of the day, basic SSR, middleware, server actions, and streaming for both routers had already started working. Next, the team completed compatibility, tests, and complex cases.
At the same time, Cloudflare does not describe the operation as a "magic button". The engineer set the architecture, set priorities, stopped unsuccessful decisions, and kept the process within strict quality gates. The cycle looked familiar to any developer: setting a task, generating code and tests, running checks, corrections and repetition. Part of the pull request review was performed by agent-based tools, but the final decisions were left to the individual.
