Vercel vs Cloudflare Workers

Fluid compute blends serverless flexibility with server-like behaviour: optimized concurrency runs multiple invocations inside a single function instance (available with the Node.js and Python runtimes), dynamic scaling adjusts resources before adding instances, background work continues after the response via waitUntil, and cold starts are mitigated through bytecode optimization and function pre-warming on production deployments. Cross-region and availability-zone failover and error isolation are built in.

Cloudflare's accounting is narrower and simpler: CPU time is what gets billed, waiting on network requests — fetch() calls, KV reads, database queries — does not count toward it, and duration carries no charge and no limit at all. The platform publishes its own reference points: the average Worker uses about 2.2 ms of CPU per request, while heavier work such as authentication, server-side rendering or large payload parsing typically lands at 10-20 ms.

Vercel Functions bill across three dimensions: Fluid Active CPU at $0.128 per hour (4 hours included), Fluid Provisioned Memory at $0.0106 per GB-hour (360 GB-hours included), and Function Invocations at $0.60 per million (1 million included). Pro is $20 per month and includes $20 of usage credit, so a small project can run inside the credit — but memory-hungry functions that stay provisioned accrue cost even while idle.

Workers Paid is a $5 monthly minimum per account and then bills two things: 10 million requests included per month with $0.30 per additional million, and 30 million CPU milliseconds included with $0.02 per additional million CPU milliseconds. Requests to static assets are free and unlimited, Cloudflare does not bill subrequests made from a Worker, and there are no data transfer or bandwidth charges — so a site serving 15 million requests with 80% static assets lands at exactly the $5 floor.

Vercel Functions allow 2 GB of memory on Hobby and 4 GB on Pro and Enterprise, with duration at 300 seconds by default, 800 seconds maximum, and an extended maximum of 1,800 seconds on Pro and Enterprise. Concurrency auto-scales to 30,000 on Hobby and Pro and 100,000 or more on Enterprise. Bundles are 250 MB uncompressed, 500 MB for Python, and large functions are supported up to 5 GB.

Every Worker is capped at 128 MB of memory on both the Free and Paid plans — the limits page lists the same figure for each tier and offers no higher option. CPU time is bounded at 5 minutes (30 seconds by default, raisable through Wrangler or the dashboard), Cron Triggers get 30 seconds, and subrequests are limited to 50 per request on Free and 10,000 on Paid. Requests are 100,000 per day on Free and unlimited on Paid.

Vercel ships official runtimes for Node.js, Bun, Python, Rust and Go, plus community runtimes, and positions its functions as powering "your framework of choice". The gap is that Fluid compute's optimized concurrency is available only with the Node.js and Python runtimes — the other runtimes do not get the instance-reuse behaviour that the pricing model assumes.

Cloudflare supports JavaScript, TypeScript, Python and Rust, and runs full-stack apps on React, Vue, Svelte, Next, Astro and React Router, deployed with a single Wrangler command and no infrastructure to configure. There is no Bun or Go runtime and no community-runtime mechanism, so a stack outside that list has no path onto the platform.