Vercel Review

Vercel is the modern-stack platform that defined the category. We host this site on it. Push-to-deploy on every PR, branch previews that feel like first-class environments, and a runtime split between Edge (millisecond-cold) and Node (everything else) — the platform's opinionated, but the opinions match how 2026 teams actually ship.

What we tested

We deployed three workloads against the Pro plan: this site (Next 16 App Router + RSC), a tools-heavy multi-region probe across iad1/fra1/hnd1 (three regions, the Pro cap — Enterprise unlocks the full five), and a static-only marketing prototype. We tracked cold-start latency, function-region pinning correctness, ISR revalidation throughput, build-cache hit rate, and the practicality of per-PR preview URLs in a small team. Performance landed at 9 / 10 — Edge handlers measured low-double-digit millisecond TTFB cold and single-digit warm across all three regions.

What we liked

The preview-per-PR model is the headline. Combined with the Vercel CLI's deploy --prebuilt flag, it removes most "works on my machine" friction. Reliability scored 9.5 / 10 — zero incidents across a quarter of testing, and the status page is honest when there is one.

What needs work

Egress and function invocations meter aggressively past the Pro plan's included quota — a single tool with caching disabled can burn a month's allowance in hours. Pricing transparency at the unit level (per-GB egress, per-million invocations) is great; the projection tools are not.