▲ hostfleet /v2
edge

Cloudflare Pages vs Vercel vs Netlify: pricing, limits, and gotchas (April 2026)

By Alex Harmon ·

If you deploy Astro, Next.js, SvelteKit, or a Vite app in 2026, you almost certainly land on Cloudflare Pages, Vercel, or Netlify. All three have free tiers. None of the free tiers mean the same thing. This article is a side-by-side of what each vendor actually publishes on their pricing pages as of April 2026, with links.

Methodology note. Every number here is sourced from the vendor’s own pricing or docs page. I have not re-benchmarked anything — you can click through every figure. Where a vendor uses fuzzy language (“generous limits”, “fair use”), I say so.

The cheat sheet

DimensionCloudflare Pages (Free)Vercel (Hobby)Netlify (Starter)
Monthly bandwidth capUnlimited (source)100 GB (source)100 GB (source)
Builds per month500 (source)6 000 build minutes (source)300 build minutes, credit-metered (source)
Concurrent builds111
Commercial use allowed on free tier?YesNo (source)Yes
Custom domainsUnlimitedUnlimitedUnlimited
Functions includedWorkers: 100k req/day (source)1M invocations, 4h Active CPU (source)125k function invocations (source)
Entry paid tierWorkers Paid $5/moPro $20 per user / moPro $19/mo

Where each one shines

Cloudflare Pages is the only one with truly unmetered bandwidth on the free tier. If you have a 200 MB wasm bundle or ship big podcast MP3s, this is the only honest free option. The catch is the 500-builds-per-month ceiling and just one concurrent build — if your CI fires on every PR commit, you can blow through the monthly allowance faster than you think. Pro tier ($20/mo as of April 2026) lifts that to 5 000 builds and 5 concurrent.

Vercel Hobby has the best Next.js ergonomics and the most generous function budget (1M invocations + 4 Active CPU hours — Active CPU means only the time the function is actually computing, not waiting). But Hobby is explicitly not for commercial use: the fair use guidelines ban selling products, running business sites, or monetizing in any way on the free tier. Pro starts at $20 per user per month, and the per-user pricing bites teams harder than it looks.

Netlify changed their pricing model in September 2025 to a credit system (300 free credits/month, metered consumption across builds, bandwidth, and functions). Read their pricing page carefully — the old “100 GB bandwidth free” framing still shows up, but consumption is now a unified credit bucket. This is flexible if you have bursty workloads but harder to reason about if you want hard caps.

The gotchas nobody mentions in the marketing

  1. Vercel Hobby forbids commercial use. Even a portfolio site running a Stripe checkout counts. See the fair use docs. People do violate this and get away with it for a while, but you can get suspended without warning.
  2. Cloudflare Pages free tier has a 25 MB per file limit on the built output (source). If you’re shipping a big video asset or a wasm binary, you’ll hit this before you hit the bandwidth cap.
  3. Netlify’s credit system is non-obvious. Build minutes, bandwidth, and serverless invocations now all drain from the same 300-credit bucket. Very hard to predict your bill if you don’t instrument it.
  4. Vercel function timeouts default to 10 seconds on Hobby, 15 on Pro (source). If you’re doing any LLM streaming, you’ll need Edge Functions or move to Pro.
  5. Cloudflare Workers paid tier is $5/mo, not $20. This is the cheapest way to get a production-ready edge function platform in 2026. Bundled with Pages, it’s hard to beat.

My recommendation as of April 2026

For a static marketing site or blog with occasional forms: Cloudflare Pages free + Workers Paid $5/mo covers 95% of what a small team needs. This is exactly what hostfleet.net runs on.

For a Next.js app where you want zero-config: Vercel Pro if it’s a real product, otherwise Cloudflare Pages with @astrojs/vercel or @cloudflare/next-on-pages.

For a team that wants to self-host eventually: Netlify remains the most portable (standard build output, not locked into one runtime), but be ready to migrate the moment your credit usage gets weird.

What this article deliberately doesn’t do

This is a pricing comparison, not a performance one. I have not re-run Lighthouse scores, Core Web Vitals, or cold-start benchmarks across the three, because (a) those results are heavily workload-dependent and (b) every other blog on the internet is already publishing made-up numbers on this. If you want to benchmark for your workload, use WebPageTest on your own deployed URLs and share the public result links — that’s the only honest way to do it.

Sources