“Run your code at the edge” was the serverless pitch of the early 2020s: put logic close to users and cut latency. In 2026 the three platforms most associated with that pitch have taken noticeably different paths. We checked each one’s official documentation and pricing on October 28, 2026 to see where code actually runs today, what limits apply, and what it costs.
At a Glance
| Spec | Cloudflare Workers | Vercel Functions | Deno Deploy |
|---|---|---|---|
| Where code runs | Cloudflare’s global network, across hundreds of locations | Configurable regions on Fluid compute; edge runtime runs nearest the request | 2 regions on the new platform, plus self-hosted regions |
| Runtime | V8 isolates | Node.js (recommended) or Edge runtime (V8) | Deno 2, with Node app support |
| Free tier | 100,000 requests/day, 10 ms CPU per invocation | Hobby: 1M invocations, 4 hrs Active CPU/month (non-commercial) | 1M requests, 10 hrs Active CPU/month |
| Entry paid plan | $5/month minimum | Pro: $20/month with $20 usage credit | Pro: $20/month |
| Edge direction in 2026 | Edge-first by design | Recommends migrating from Edge to Node.js | Fewer regions than Deploy Classic, which shut down July 20, 2026 |
Short Verdict
If “edge” is the requirement, meaning code that runs near every user by default, Cloudflare Workers is the platform still built around it, with a $5 paid plan and CPU-time billing that does not charge for time spent waiting on network calls. Vercel now steers developers away from its Edge runtime and toward Node.js on Fluid compute, which favors running functions near your data rather than near your users. The new Deno Deploy runs in two regions, so it is better viewed as a serverless platform than an edge network. Pick based on where your data lives: if your database sits in one region, compute next to it usually beats compute next to the user.
What “Edge” Means Here
Edge computing means processing a request in a location close to the user instead of in one central region. It shortens the network round trip for work that does not need a central database, such as redirects, authentication checks, personalization from cookies, A/B test assignment, and caching logic. It helps much less when every request must read from a single-region database, because the edge function then makes its own long round trip to that database.
Cloudflare Workers
Workers run on Cloudflare’s global network, which the company describes as thousands of machines across hundreds of locations, and a request arriving at any Cloudflare data center can invoke your Worker. Instead of a container or virtual machine per function, each Worker runs in a V8 isolate inside an existing runtime process. Cloudflare says an isolate can start around a hundred times faster than a Node process in a container or VM and uses an order of magnitude less memory at startup, which is why Workers avoid the classic cold-start penalty. Isolates can be evicted at any time, so global mutable state should not be relied on.
| Plan | Requests | CPU time | Price |
|---|---|---|---|
| Free | 100,000/day | 10 ms per invocation | $0 |
| Paid (Standard) | 10M/month included, then $0.30 per million | 30M CPU-ms/month included, then $0.02 per million CPU-ms; up to 5 minutes CPU per request (30 s default) | $5/month minimum |
Cloudflare does not charge for duration, only CPU time, so a Worker that spends most of its time waiting on a fetch is billed only for the milliseconds it actually computes.
Vercel: From Edge Runtime to Fluid Compute
Vercel’s Edge runtime still exists and, by default, executes in the region closest to the incoming request. But Vercel’s own documentation now recommends migrating from Edge to Node.js “for improved performance and reliability,” notes that both runtimes run on Fluid compute with Active CPU pricing, and states that starting in Next.js 16.3, setting runtime = 'edge' is no longer supported. The Edge runtime also has real constraints: a subset of Web and Node.js APIs, no filesystem access, no eval, a requirement to begin responding within 25 seconds, and code size limits of 1 MB (Hobby), 2 MB (Pro), and 4 MB (Enterprise) after compression.
| Resource | Hobby (included) | Pro (beyond included credit) |
|---|---|---|
| Active CPU | 4 hours/month | From $0.128/hour |
| Provisioned memory | 360 GB-hours/month | From $0.0106/GB-hour |
| Function invocations | 1M/month | From $0.60 per million |
Hobby is for personal, non-commercial use and cannot buy extra usage. Pro costs $20 a month, includes $20 of usage credit, and is pay-as-you-go beyond it.
Deno Deploy: Rebuilt, and Fewer Regions
Deno rebuilt Deploy on a new execution environment using Deno 2.0 that runs both Deno and Node apps, with first-class support for Next.js, Astro, and SvelteKit. Its documentation lists 2 regions for the new platform, compared with 6 for Deploy Classic, plus the option of self-hosted regions. Deno announced that Deploy Classic and the subhosting v1 API would shut down on July 20, 2026.
| Plan | Price | Requests/month | Active CPU | Egress |
|---|---|---|---|---|
| Free | $0 | 1M | 10 hours | 20 GiB |
| Pro | $20/month | 5M, then $2 per million | 50 hours, then $0.10/hour | 200 GiB, then $0.20/GiB |
| Builder | $200/month | 25M, then $2 per million | 500 hours, then $0.10/hour | 2,000 GiB, then $0.20/GiB |
| Enterprise | Custom | Custom | Custom | Custom; 99.95% SLA |
Pros and Cons by Platform
- Cloudflare Workers pros: genuinely global by default, isolate-based startup, cheap $5 entry plan, no charge for waiting time. Cons: isolate runtime is not full Node.js, and the free tier’s 10 ms CPU cap is tight for heavier work.
- Vercel pros: tight integration with Next.js and Fluid compute, configurable regions near your data. Cons: Edge runtime is being de-emphasized and is unsupported in Next.js 16.3+, and Hobby is non-commercial only.
- Deno Deploy pros: generous free CPU allowance, Deno and Node apps, self-hosted regions, clear usage pricing. Cons: only 2 managed regions on the new platform, and teams on Deploy Classic had to migrate.
Bottom Line
The industry has quietly split “edge” in two. Cloudflare Workers still runs your code near the user everywhere, which is ideal for request-level logic that does not need a central database. Vercel and Deno have moved toward regional serverless compute placed near your data, which suits typical database-backed apps better. Decide which latency you are actually fighting, user-to-server or server-to-database, and the platform choice follows.
Frequently Asked Questions
- Is Vercel’s Edge runtime deprecated? Vercel recommends migrating from Edge to Node.js, and setting
runtime = 'edge'is no longer supported starting in Next.js 16.3. - How much does Cloudflare Workers cost? Free for 100,000 requests a day; the paid plan starts at $5 a month with 10 million requests and 30 million CPU milliseconds included.
- Is Deno Deploy an edge platform? The new Deno Deploy lists 2 managed regions, so it behaves more like regional serverless; Deploy Classic, which had 6 regions, shut down on July 20, 2026.
- When does edge computing not help? When every request must reach a single-region database, since the edge function then adds its own long round trip.