Vercel vs Netlify: Pricing, Limits and a TTFB Check

Personal site: Vercel Hobby. A site that takes payment: Netlify Free or Vercel Pro. Next.js: Vercel. Static traffic past either free cap: Cloudflare Pages.

By David · Published November 8, 2026 · 16 min read

Vercel vs Netlify is a host choice for a small site or a SaaS that deploys from Git. People also search netlify vs vercel. It lives on the backend and stack hub. The reader wants the free cap, the paid price, and a deploy of the same site.

I did not create a Vercel account or a Netlify account, and I did not deploy one Astro site to both free tiers. There is no build time from either platform on this page. The timing below is 20 new requests from one VM in Ashburn, Virginia, to each vendor’s public Astro starter. The prices are the vendor pages on November 8, 2026. Where a page does not publish a figure, the cell says so.

Pricing and free tier limits

Vercel Hobby is $0. The Hobby plan includes 100 GB of Fast Data Transfer, 10 GB of Fast Origin Transfer, 1,000,000 CDN requests, 4 CPU-hours of active CPU, 360 GB-hours of provisioned memory, and 1,000,000 function invocations. The same page says there is no billing cycle on Hobby. If you pass a usage limit, you wait until 30 days have passed before you can use that feature again. The account plans page says Hobby plans are paused when they exceed the included free tier usage. Vercel’s pages say the prices are in USD and exclude VAT and GST.

The fair use guidelines restrict Hobby to non-commercial, personal use. Commercial use needs Pro or Enterprise. The page defines commercial use as a deployment for the financial gain of anyone involved in producing the project, including a paid employee or consultant who writes the code. The examples include taking a payment, advertising a product or service, being paid to create or host the site, a site whose primary purpose is affiliate links, and ads, including AdSense. The same page says asking for donations is not commercial use.

Netlify Free is $0 with a 300 credit limit, on the pricing page. The credit-based plans page says new accounts from September 4, 2025 use credit-based plans, and Free is 300 credits a month with a hard limit and no auto recharge. The how credits work page says that when the credit balance is used up, every web project is paused and visitors get a “Site not available” page. Unused monthly credits do not roll over on Free.

How credits work prices the meters that spend those credits. A production deploy is 15 credits, and the page says build minutes are no longer calculated. Deploy previews and branch deploys are 0 credits. Bandwidth is 20 credits per GB. Web requests are 2 credits per 10,000 requests, and that meter includes page views, API calls, redirects, function requests, and asset requests. Compute is 10 credits per GB-hour. The pricing page prints the same four rates.

The three Netlify “if every credit” rows are exclusive. They are not a bundle. One production deploy is 15 credits. 300 minus 15 is 285. 285 divided by 20 is 14.25 GB of bandwidth, with no requests and no compute in that arithmetic. Twenty production deploys, 300 divided by 15, spend the month and leave 0 GB. I did not run a site, so this is the rate, not a bill.

The November 12, 2024 announcement, Introducing Netlify’s Free plan, says you can deploy commercial projects, personal sites, or other work you want to share. The same post lists 100 GB of bandwidth and 300 build minutes. Those two numbers are the older meter. They are not the 300 credit plan in the table above. I am using the announcement for the commercial sentence, and the November 8 pages for the cap.

The paid plans

Vercel Pro, on the Pro plan page, is a $20 a month platform fee. The fee includes one deploying team seat and $20 of usage credit. The credit applies across managed infrastructure, expires at the end of the month, and usage past it is billed on demand. The pricing page prices an additional paid seat at $20 a month. Flat Rate CDN, on the Pro plan page, includes the lowest tier at no extra cost: 1,000,000 CDN requests and 1 TB of data transfer a month. I am not treating that 1 TB as something the $20 credit has to buy. The page lists the tier as included.

Netlify Personal, on the pricing page, is $9 a month and 1,000 credits. Netlify Pro is $20 a month, 3,000 credits, and unlimited members. Personal can buy 500 more credits for $5. Pro can buy 1,500 more credits for $10. Free cannot. At the Pro pack rate the pricing page prints, $10 for 1,500 credits, a production deploy is $0.10, bandwidth is $0.13 per GB, and compute is $0.07 per GB-hour. That dollar column is their conversion of the credit rates. It is not a second price list.

• 100 GB

  Vercel Hobby Fast Data Transfer

• 15 GB

  Netlify Free, if all 300 credits are bandwidth

• $20

  Vercel Pro fee and Netlify Pro price

• $9

  Netlify Personal

A small SaaS that takes a payment cannot stay on Vercel Hobby. The fair use page says so. The same app can start on Netlify Free until the 300 credits run out, or on Vercel Pro at $20. The database under that app is a different bill. Supabase vs Firebase prices one workload at 1,000, 10,000, and 50,000 users. This page does not.

Build and deploy speed

I did not time a build. The limits below are the published caps, not a stopwatch.

On Vercel, the limits page sets the build step to 45 minutes on Hobby and on Pro. A build that reaches 45 minutes fails. Hobby allows 1 concurrent deployment and 100 deployments created per day. A CLI upload of source files is capped at 100 MB on Hobby and 1 GB on Pro. The Hobby comparison table lists the build machine as 2 vCPUs, 8 GB of memory, and 32 GB of disk. The same limits page says a Hobby team cannot connect a project to a Git repository owned by a Git organization.

On Netlify, a production deploy costs 15 credits and build minutes are no longer a separate meter. Previews are free. The credit-based plans page says Free includes 1 concurrent build, so only one build runs at a time. I did not click a Git deploy on either host, so this page does not name a button I did not see and does not report a build duration.

The measurement I could run is the response time of a page that is already deployed. The Vercel Astro boilerplate names https://astro-template.vercel.app as the live example. Netlify’s own post, Make and Deploy an Astro Project in Minutes, links https://astro-quickstart.netlify.app. I opened https://astro-platform-starter.netlify.app as well. Its title was “bye”, so I did not time it.

What we tested

Checked November 8, 2026

The clock started at 2026-11-08T09:06:28.997Z and the last request finished at 2026-11-08T09:06:32.551Z. That window includes two warmup requests per URL, which are not in the table. Then 20 measured requests, interleaved, Vercel then Netlify, one new curl process each time so the connection was not reused. curl 8.5.0, HTTP/2, with compressed responses. The User-Agent was shipmunks-ttfb/1.0. Every measured response was HTTP 200.

ipinfo.io, checked just before the run, reported Ashburn, Virginia, Amazon AS14618, address 34.199.12.19. A later check from the same VM was 3.226.203.3, the same city. Vercel’s x-vercel-id started with iad1 on every saved header. Netlify’s headers did not name a city. The Netlify addresses curl reported were 18.208.88.157 and 98.84.224.111.

TTFB is curl’s time_starttransfer. Total time is time_total. Both include DNS, TCP, and TLS, because each request was a new process. “After TLS” is time_starttransfer minus time_appconnect, and the median is the median of that difference, not the difference of the medians. For an even count, the median is the mean of the two middle sorted samples.

A probe before the timed run, one request each, with no Accept-Encoding, returned content-length 4,825 for the Vercel URL (title “Welcome to Astro.”) and 15,713 for the Netlify URL (title “Astro Quickstart”). https://astro.vercel.app returned the same etag and the same 4,825 bytes as the template URL on that probe. The timed run sent Accept-Encoding and downloaded brotli: 1,742 bytes on every Vercel sample, and 4,116 to 4,134 bytes on Netlify.

Vercel was x-vercel-cache: HIT on all 20. age was 250162 to 250164 seconds. last-modified was Thu, 05 Nov 2026 11:37:07 GMT. A unique query string on five later requests was still HIT. Netlify’s cache-status was a miss on 15 of the 20 (fwd=miss) and a hit on 5. The pages are different sizes. This is not one Astro site deployed twice, and it is not a cold origin on Vercel.

Medians from the 20 measured requests, November 8, 2026. Times are milliseconds.

Group	n	Median TTFB	Min	Max	Median total	Median after TLS
Vercel template, all hits	20	49.3505	45.622	274.274	49.436	12.644
Netlify quickstart, all samples	20	65.754	30.091	92.895	65.960	38.341
Netlify, hits only	5	31.593	30.091	47.344	31.717	4.885
Netlify, misses only	15	67.595	51.113	92.895	67.733	39.918

The Vercel median is the mean of the 10th and 11th sorted samples, 49.307 ms and 49.394 ms, which is 49.3505 ms. The chart rounds that to 49.351 ms. The 274.274 ms sample is real. Its TLS time was 264.039 ms, so the stall was in the handshake. The other 19 Vercel samples were 68.352 ms or faster. Total time sat within 1 ms of TTFB on both hosts, because the brotli bodies were a few kilobytes.

The five Netlify hits were faster than the Vercel hits. The fifteen misses were slower. The combined Netlify median, 65.754 ms, sits with the misses because 15 of 20 samples were misses. I am not calling either host faster. The cache state and the page are different.

Five more requests used a unique query string. They are not in the medians above. Vercel stayed HIT, TTFB from 46.841 ms to 54.730 ms, median 50.415 ms. Netlify was four misses and one hit, TTFB from 30.029 ms to 79.198 ms, median 67.212 ms.

• Vercel, 20 hits 49.351 ms
• Netlify, 5 hits 31.593 ms
• Netlify, 15 misses 67.595 ms

Sample	Vercel TTFB	Vercel total	Netlify TTFB	Netlify total	Netlify cache
1	50.852	50.937	66.492	66.710	miss
2	47.394	47.481	64.238	64.449	miss
3	49.694	49.779	67.144	67.411	miss
4	49.147	49.238	67.595	67.733	miss
5	51.081	51.181	63.338	63.794	miss
6	49.307	49.391	77.366	77.636	miss
7	68.352	68.436	51.113	51.242	miss
8	47.793	47.903	64.118	64.312	miss
9	49.241	49.342	74.738	75.039	miss
10	45.622	45.726	83.795	84.495	miss
11	46.940	47.033	30.091	30.220	hit
12	49.851	49.949	65.016	65.210	miss
13	48.517	48.601	31.866	31.986	hit
14	274.274	274.362	82.480	82.724	miss
15	47.321	47.420	87.304	87.490	miss
16	59.140	59.241	30.724	30.846	hit
17	52.827	52.929	80.067	80.240	miss
18	49.394	49.481	92.895	93.021	miss
19	50.065	50.152	31.593	31.717	hit
20	48.641	48.733	47.344	47.804	hit

Serverless and edge functions

Vercel Functions, on the limits page for Fluid compute, allow 2 GB of memory on Hobby and a maximum duration of 300 seconds, which is also the default. The uncompressed size limit is 250 MB, or 500 MB for Python. The default region is iad1, which you can change. Pro and Enterprise can set more than one region. The same page says an Edge runtime function must begin sending a response within 25 seconds, and can keep streaming for up to 300 seconds. A separate line, on the fair use page, caps middleware that uses the Edge runtime at 50 ms of CPU on average. That 50 ms is CPU time, and time spent waiting on the network does not count. It is not the 300 second duration.

Netlify Functions, on the configuration page, default to region cmh (US East, Ohio), 1024 MB of memory, and a synchronous limit of 60 seconds. Background functions can run 15 minutes. The buffered payload limit is 6 MB. Raising memory or vCPU is listed as a Credit-based Pro and Enterprise feature, from 1024 MB to 4096 MB. Free stays at the 1024 MB default.

Netlify Edge Functions, on the limits page, allow a 20 MB bundle after compression, 512 MB of memory for the set of deployed edge functions, 50 ms of CPU per request, and a 40 second response-header timeout. CPU time does not include waiting on the network. The page says the monthly invocation count depends on the team plan and points you at the usage dashboard. It does not print a free-plan number, so this post does not invent one.

The two 50 ms figures are both CPU caps, on different products. Vercel’s is the fair-use average for Edge middleware. Netlify’s is the per-request cap for Edge Functions. I did not deploy a function to either host, so there is no duration from a handler I ran.

If the thing you are choosing is the handler itself, the empty-route rates on one VM are in Hono vs Express. That page did not deploy Hono to Vercel or to Netlify. The Vercel frameworks page lists Hono under backends. Listing it is not a timing result.

Framework support

Vercel builds Next.js, and the Hobby and Pro pages treat a Git push as the deploy. The frameworks page, read November 8, 2026, also lists Astro with the frontends. The Astro boilerplate says the starter deploys with zero configuration. The live example is the URL in the test above.

Netlify’s Next.js page, last updated February 4, 2026, says the adapter supports Next.js from version 13.5, and that leaving the adapter unpinned updates it on each build. The support table marks App Router, server rendering, incremental static regeneration, static generation, React Server Components, Server Actions, and image optimization as supported. The limitations on that page are the ones that matter if you are leaving Vercel:

• Server-rendered pages that use the Edge runtime run in the functions region on the Node.js runtime, not at the edge. The page says no functionality is missing.
• A rewrite in the Next.js config cannot point at a static file in public.
• Headers and redirects are evaluated after middleware, which differs from stand-alone Next.js.
• Node.js middleware does not support C++ addons or the filesystem.

For a static Astro site, both hosts have a public starter, and both starters answered HTTP 200 from this VM. For a Next.js app, start on Vercel. Move it to Netlify only after those four limits are acceptable for the app you have. I did not port an app, so this is the docs, not a migration I finished. Sign-in on that Next.js app is a separate vendor. Clerk or Auth0 on a Next.js app is the pricing, from their pages, not from a login I ran.

When to pick Cloudflare Pages instead

Cloudflare Pages is the static host that does not spend the request on the meters above. The Pages functions pricing page, last updated September 8, 2026, says requests to static assets are free and unlimited on free and paid plans, as long as the request does not invoke Functions. Functions count toward the Workers quota. The page’s own example is 50,000 Functions requests plus 50,000 Workers requests against 100,000 a day on the Free plan, resetting at midnight UTC.

The Pages limits page lists 500 builds a month on Free, one build at a time, a 20 minute build timeout, and 20,000 files. That is a build count, not Netlify’s 15 credits and not Vercel’s 100 deployments a day. The feature split between Workers and Pages, including what a new project should use, is Cloudflare Workers vs Pages. This page did not re-run that dry run.

A long-running process with Postgres on the same host is a different product. The worked month for that shape is Railway vs Render. Once the site is on a host and the domain answers, the order for getting it into Google is how to get a new website indexed.