Cloudflare just released Containers: here's everything you need to know

Cloudflare Containers let you run any Docker image on Cloudflare's 300-plus edge locations. You control them with a few lines of JavaScript in a Worker, they scale to zero, and you're billed in 10 ms slices while they're awake.

They sit in the gap between:

Cloudflare Containers bring the edge reach and pay-for-use pricing of Workers to workloads that need a full Linux sandbox.

Why would I care?

• Native binaries or full FS, so you can run FFmpeg, Pandas, or AI toolchains.
• Languages beyond JS or Wasm, such as Go, Rust, Python, Java, Ruby, or anything your Dockerfile holds.
• Bigger resource envelope, with up to 4 GiB RAM and half a vCPU per instance (larger sizes are planned).
• Per-tenant state, with one container per Durable-Object ID for sticky sessions.
• Burst-heavy jobs, such as cron, code evaluation, or on-demand video export.

If your code sleeps a lot, scaling to zero is better than paying for an always-on container (whether that is sliplane, a VPS, or a managed dyno).

How it works

# 1. Scaffold + deploy
npm create cloudflare@latest -- --template=cloudflare/templates/containers-template
wrangler deploy

// 2. Route requests in your Worker
import { Container, getRandom } from "@cloudflare/containers";

class API extends Container {
  defaultPort = 8080;
  sleepAfter  = "10m";
}

export default {
  async fetch(req, env) {
    const instance = getRandom(env.API, 3);   // simple round-robin helper
    return instance.fetch(req);
  },
};

The first hit is a cold-start (about 2 to 3 seconds in beta). After that, the container stays warm until it is idle for the duration set in sleepAfter.

Under the hood, each container is coupled to a Durable Object that handles lifecycle and routing. There is no YAML, no nodes, just code.

Pricing snapshot

Instance sizes in beta are dev (256 MiB), basic (1 GiB), and standard (4 GiB). Larger sizes are coming.

Assume a "standard" instance (4 GiB RAM, half a vCPU, 4 GB disk) that runs 24 × 7 for a 30-day month and ships 2 TB of traffic. This is a workload better suited to an always-on PaaS.

Variable total: about $77.44 per month. Add the $5 Workers Paid subscription, and the total is about $82.44 all-in.

A comparable always-on PaaS instance (such as sliplane or a small VPS) might cost $7 to $15 per month flat, so for high-utilisation, bandwidth-heavy services, Cloudflare Containers can be five to ten times more expensive.

Rule of thumb: workloads that idle most of the day tend to cost less on Containers. Steady-state, high-utilisation services can still be cheaper on an always-on host like sliplane.

Current beta limits

1. Manual scaling, you call get(id). Autoscale and latency routing are planned.
2. Ephemeral disk, so you get a fresh FS after each sleep.
3. 40 GiB RAM and 20 vCPU account cap (temporary).
4. Linux/amd64 only, no ARM support yet.
5. No inbound TCP or UDP, since everything is proxied through a Worker HTTP call.

When to pick Containers vs. an always-on PaaS

A mixed model often wins. You can run your persistent database on sliplane (or similar), bursty compute on Cloudflare Containers, and connect them with a Worker.

Takeaway

Cloudflare just introduced what is essentially serverless Fargate at the edge: Docker images, millisecond billing, global points of presence, and no cluster busywork. If Workers' V8 box ever felt cramped, or your always-on container spends most of its time idle, try spinning up a beta Container and see what the edge can do.

Happy hacking!

Cheers,

Jonas, Co-Founder of sliplane