Next.js 16.4 leaves existing apps on old caching until they opt in
Next.js 16.4 recommends Cache Components but only enables them for new projects. Here is what existing apps must do, and what benchmarks show.
Next.js 16.4, released October 6, 2026, makes Cache Components the recommended caching model for every app, but it turns them on by default only for new projects. Existing apps keep their current behavior until they set a flag, according to an independent migration guide on DEV Community and Roboto Studio's release write-up. Vercel's release post is the primary source, though its full text was not readable for this report. Roboto's post is dated October 6 in its text, and a search listing showing November 2 appears to reflect a later update; EasySpawn also gives October 6.
The upgrade is therefore two jobs. Installing 16.4 is a version bump. Adopting Cache Components changes caching from per-route static-or-dynamic decisions to per-component ones. Roboto says the model becomes the default in Next.js 17.
The opt-in is cacheComponents: true in next.config.ts (Roboto and EasySpawn also pair it with partialPrefetching: true). Functions or components you want cached get a use cache directive, a lifetime via cacheLife and tags via cacheTag. Code reading cookies, headers or search params renders at request time behind a Suspense boundary. The DEV guide calls this a contract change: teams must decide how stale each piece of data may be.
The release also adds an experimental agent mode, run as npx next@canary upgrade --agent=latest. The DEV guide advises running it on a branch, then npm run build, and keeping freshness and per-visitor data decisions with humans. Its suggested rollout is to ship the version bump with the flag off, then enable the app-wide config on a branch and add and test cache boundaries route group by route group. It also proposes a staging check requesting one page with two different cart cookies; commenters add that tenant leaks need users from different organizations. Roboto reported the agent command failed when pointed at a subdirectory with a relative import in its config, but worked from inside the app directory.
Performance figures come from Roboto, not Vercel. On a 1,555-page site, the first build after installing dependencies fell from 83.9s on 16.3 to 49.1s on 16.4, and Roboto cautions that it had only one such 16.4 run. Later cold builds took 25.3–30.2s versus 30.2s on 16.3, and warm repeat builds were essentially unchanged (20.6s versus 20.9s). Roboto measured about 5% less uncompressed JavaScript on disk across all chunks, and a 282MB to 239MB smaller .next/cache. Vercel's own numbers were not verified.
Security: Vercel's security page listing and Netlify's changelog describe an October 14 out-of-band update with fixes in 16.4.1 and 15.5.28. Netlify says the critical use cache issue affects 16.x apps using Cache Components that return typed arrays. The Hacker News reports two critical remote code execution flaws, in caching and SVG processing. Advisory pages could not be read, so check the affected-version ranges in the GitHub advisories before deciding which release to install.
Comments ()