Next.js Has Two Supported Lanes. Your Migration Has to Pick One.
The September patches shipped to both 16.3.8 and 15.5.27. That pair of labels is a fork in your migration scope - and the cheaper lane is not always the older one.
Two version numbers, one batch of fixes
On 30 September 2026 the Next.js team published the September security release. The fixes went out in two places at once: v16.3.8, labelled Active LTS, and v15.5.27, labelled Maintenance LTS. One batch of advisories, two supported lanes, both patched the same day.
If you are holding a quote to move a Lovable prototype onto Next.js, that pair of labels is a decision nobody has put in front of you yet. The quote says Next.js. It does not say which lane. And the lane is not a version-number detail: it decides which router you write against, which bundler builds your app, which features you are allowed to turn on, and how much of the September advisory list applies to you at all.
Use the latest answers a different question
Next.js 16.3 landed on 3 August 2026. The release highlights were up to 90% less memory in development, faster builds, and Instant Navigations - wins that are real and that a founder feels immediately in the dev loop.
Those are good reasons to be on 16.3. They are not, on their own, reasons to put a two-week-old prototype on the newest lane. New capability is new surface area, and the September advisories show exactly where that surface is.
Your exposure is a function of flags, not of your version number
Read the September release as a list of conditions rather than a list of versions. Almost every item names the configuration that makes you vulnerable - and the configuration that makes you safe.
| September 2026 advisory | You are affected if | You are not affected if |
|---|---|---|
Image Optimization SSRF (CVE-2026-94483, High) | You have images.remotePatterns configured | No remotePatterns are configured |
SSG and ISR cache poisoning (CVE-2026-94543) | You self-host, use the Pages Router, and serve SSG or ISR pages | You deploy on Vercel |
Metadata image route disclosure (CVE-2026-94485) | Your App Router app is built with webpack | Your app is built with Turbopack |
Draft Mode leak through a pending use cache fill (CVE-2026-94544) | You enable Cache Components and serve Draft Mode previews | Cache Components is off |
Dev server MCP endpoint disclosure (CVE-2026-94486, Low) | A developer is running next dev | It is a production deployment |
Not one of those rows is settled by the number after next@. Every one of them is settled by a decision somebody makes during the migration - usually in the first week, usually without writing it down.
What the lane actually changes in a quote
Four line items move when the lane moves.
Router. The Pages Router appears by name in the self-hosted cache-poisoning advisory. An App Router migration costs more up front and takes you off that particular row.
Bundler. One of September's medium-severity disclosures applies to webpack builds and explicitly does not apply to Turbopack builds. Which bundler your migration lands on is a scoping decision with a security consequence attached, not a preference.
Hosting. The phrase applications deployed on Vercel are not affected appears in one advisory. Self-hosting is a legitimate choice with a cost, and part of that cost is that some upstream mitigations stop being free.
Caching. Two of September's items apply only when Cache Components, or experimental.useCache, is enabled. A prototype needs neither. An app that switches them on in month two inherits both rows.
The cadence is the line item people forget
Five weeks of 2026, on both lanes at once:
- 25 August - 16.3.3 and 15.5.24, addressing two Critical vulnerabilities.
- 22 September - an out-of-band update to 16.3.6 and 15.5.26 for a critical upstream issue.
- 30 September - 16.3.8 and 15.5.27, covering one High, five Medium and one Low.
Three patch events in thirty-six days. Whichever lane you pick, that is the rhythm you are agreeing to. A migration that ends at handover with nobody owning the upgrade has not finished - it has moved the problem onto a calendar nobody is reading.
Supabase has dates in the same window
If the migration also owns the backend, three more items land inside the same quarter, all published on the Supabase changelog:
- 1 December 2026. The Hono, H3, Elysia and NestJS adapters are removed from
@supabase/server. They still work today. The changelog is explicit that they are removed on that date, and points at single-file bridges that compose middleware entries instead. - 31 January 2027.
supabase-jsbegins requiring TypeScript 5.0, dropping support for 4.7 through 4.9. - Already in effect. Supabase Middleware reached 1.0 on 30 September 2026 and now follows semantic versioning, so breaking changes ship only in a new major. That is a promise you can plan a dependency range around, and it only helps you if somebody sets the range.
None of these is dramatic. All of them are dated. The difference between a maintained app and a stalled one is usually just whether somebody wrote the dates down.
How we scope this
Which lane your app lands on is the first question we answer, not the last. Our Strategy Audit is $499, one-time: a one-hour scoping call, a written technical roadmap, a feature prioritization matrix, a stack and architecture recommendation, and a timeline and cost estimate - and the $499 is credited toward any build. The lane, the router, the bundler and the hosting decision are all settled there, in writing, before a line of migration code is written.
From there, the Publishable MVP is $4,999, one-time, and ships ready to publish: landing page, auth, Stripe payments, user dashboard, admin panel, analytics, blog, SEO system, production deployment, monitoring, and documentation plus handoff, with 90-day support until your first paying customer. The Founding Client tier is $2,999, one-time, with a 14-day build guarantee and a full refund if the deadline is missed. Both carry our revenue guarantee: first paying customers within 90 days of launch, or a full refund.
We have completed 80 jobs on Upwork, and the pattern is consistent. The apps that stay healthy are the ones where somebody picked the lane on purpose.