跪拜 Guibai
← All articles
Vue.js

Vue 2026: Core Rewritten, API Frozen, and Upgrading Is Nearly Free

By 用户451419429322 ·
Read original on juejin.cn ↗ Google Translate ↗ Alt translation

The 2026 Vue upgrade path proves that a framework can modernize its internals without breaking the API surface, but the real risk has shifted to adoption sequence: enabling Vapor too early or mixing it with the reactivity upgrade makes regressions impossible to attribute. For teams on Vue 3.5, the cheapest win is the build layer, not the renderer.

Summary

As of November 2026, Vue 3.6 is still in RC, but the underlying stack has already been fully replaced: the reactivity engine is rewritten on alien-signals with a push-pull two-phase model, Vapor Mode moves rendering from virtual DOM diffing to compile-time imperative DOM directives, and Vite 8 swaps esbuild and Rollup for a single Rust-based bundler, Rolldown. None of these changes require touching business code. Vue 4.0 has also been announced explicitly to avoid breaking API changes, so this round of investment is safe.

Measured on a minimal demo, a full Vapor app's gzip size drops from about 45.4 kB to 18.0 kB, a 60% reduction, because the vDOM runtime is tree-shaken. Hybrid mode, however, is slightly larger than pure vDOM, adding about 0.5 kB, because it bundles both runtimes and an interop plugin. That's why the official advice is to pilot Vapor on one or two performance-sensitive pages in existing apps, not to migrate wholesale. The reactivity rewrite, meanwhile, is the only change that takes effect on upgrade without opt-in, so complex watcher and effect chains need a dedicated regression pass before enabling Vapor on top.

The build layer delivers the biggest and safest win: Vite 8's Rolldown offers 10-30x build speedups on large codebases, with reported cases like Linear dropping from 46s to 6s. The toolchain also got a unified `vp` command via Vite+ 1.0, and VoidZero's Cloudflare acquisition came with a public MIT commitment for all its projects. The recommended upgrade sequence is Vite 8 first, then 3.6 stable without Vapor, then targeted Vapor pilots. Performance numbers from synthetic demos should be validated against your own pages.

Takeaways
— Vue 3.6 rewrites the reactivity core on alien-signals using a push-pull model and a bidirectional linked list, cutting baseline memory use by about 14%.
— Vapor Mode is opt-in per component and compiles templates to direct DOM update directives, eliminating VNode generation and diffing at runtime.
— A minimal demo shows a full Vapor app at 18.0 kB gzip versus 45.4 kB for pure vDOM, a 60% reduction; hybrid mode is 45.9 kB, slightly larger than the vDOM baseline.
— Vapor does not support Options API, getCurrentInstance(), globalProperties, v-memo, $el/$props proxies, custom directives, Suspense (except functional equivalence), and element lifecycle events.
— The reactivity rewrite is globally effective on upgrade to 3.6 without opt-in; complex watcher and effect chains may see timing changes and need regression testing.
— Vite 8 unifies dev and build on Rolldown, giving 10-30x build speedups on large codebases; install size grows by about 15MB for CI planning.
— Vite+ 1.0 introduces a single `vp` command for dev, build, test, lint, format, and monorepo tasks; `vp check` runs all static checks in one pass.
— VoidZero joined Cloudflare in June 2026 and committed to keeping Vite, Vitest, Rolldown, Oxc, and Vite+ open source under MIT.
— Nuxt 3 is EOL; Nuxt 4.6 supports Vapor interop, and most Nuxt 5 capabilities are already opt-in in 4.6.
— Pinia 3 drops Vue 2 support; Pinia Colada covers server state with caching, deduplication, background revalidation, and optimistic updates.
— The recommended upgrade order for existing projects: Vite 8 first, then 3.6 without Vapor, then pilot Vapor on 1-2 genuinely bottlenecked pages.
— Performance claims like 97% rendering improvement come from synthetic demos; measure on your own pages before trusting them.
Conclusions

The Vue team's strategy of burying all aggressive changes inside the compiler and runtime means the real cost of adoption is shifting from code changes to testing effort: the reactivity rewrite changes dependency timing silently, which is harder to debug than a compile error.

The hybrid-mode size penalty of 0.5 kB exposes a tradeoff the marketing usually hides: Vapor's benefits are only clean when you go all-in, which explains why the official guidance is page-level pilots or full-Vapor new apps.

Evan You's rationale for Vue 4.0 — keeping APIs stable because AI training data has fossilized them — signals that framework design is now partly driven by the need to keep AI code generators accurate, a new constraint on language evolution.

The satisfaction gap (84% Vue vs 72% React) persists even as React dominates usage and downloads, suggesting that 'better to use' and 'more popular' are diverging metrics, and that ecosystem size often overrides developer preference.

The 10-30x build speedup from Rolldown is the kind of improvement that shows up in daily workflow, not just benchmarks, which is why the article advises upgrading the build layer first.

Concepts & terms
Vapor Mode
A compile-time rendering strategy for Vue where templates are analyzed at build time and translated into direct DOM update instructions, eliminating the virtual DOM and diffing at runtime. It is opt-in per component and requires the Composition API.
alien-signals
The reactivity library that Vue 3.6's reactivity core is rewritten on. It uses a push-pull two-phase propagation model with a bidirectional linked list for dependencies, improving memory usage and stale dependency cleanup.
Rolldown
A Rust-based bundler that becomes the single bundler in Vite 8, replacing esbuild in dev and Rollup in build. It aims for plugin compatibility while delivering significant build speedups.
Pinia Colada
A server-state management solution for Vue, similar to TanStack Query, providing caching, request deduplication, background revalidation, and optimistic updates for async data.
push-pull propagation
A reactivity model where changes to signals only mark dependent computed values as dirty (push), and computed values recompute only when read (pull), avoiding unnecessary work and improving cache efficiency.
Source: juejin.cn ↗ Google Translate ↗ Backup ↗