Vercel Sandbox vs Northflank: sandbox pricing, isolation and limits

Which is better, Vercel Sandbox or Northflank?

It depends on how busy your sandboxes are. For agents that mostly wait on a model, Runtime has the lowest bill: $0.64 for the job below, against $1.11 on Northflank and $2.12 on Vercel. Its pause and forks also keep memory. Pick Vercel Sandbox if the app is on Vercel, with up to 10,000 sandboxes on Pro. Pick Northflank for CPU-bound jobs that keep every core busy, where it is cheapest, or for sandboxes inside your own cloud.

At a glance

Vercel is priced at its default region, iad1.

	Vercel Sandbox	Northflank	Runtime
Isolation	Firecracker microVM, own kernel	microVMs for CPU workloads, gVisor for GPU workloads	Firecracker microVM, own kernel
CPU	$0.128 per active CPU-hour	$0.01667 per allocated vCPU-hour	$0.025 per vCPU-hour of measured CPU, with a small floor
Memory	$0.0212 per provisioned GB-hour; 2 GB per vCPU	$0.00833 per GB-hour	$0.0075 per reserved GiB-hour
Plan fee	Hobby allowance free; usage beyond it needs Pro	None published for compute	None; prepaid credit from $10
Free start	5 active CPU-hours and 420 GB-hours a month on Hobby	2 services, 2 jobs and 1 database free	100 free hours of a 1 GB machine in the first 30 days, no card
Session length	45 minutes on Hobby, 24 hours on Pro, per session	No time limit	No time limit: runs while it works, pauses itself after 300 seconds idle
Where it runs	19 Vercel regions	Northflank's regions, or your AWS, GCP, Azure or Oracle VPC	Dedicated servers Runtime operates
What you create	A sandbox, in a Vercel project	A service with a deployment plan	A sandbox, in one call

Why is Northflank cheaper even though Vercel bills only active CPU?

Because the rates are far apart. Vercel's active CPU costs $0.128 an hour. Northflank's allocated vCPU costs $0.01667, under a seventh as much. Vercel bills only the CPU the code uses, but even an agent busy a sixth of the time pays more on Vercel than for a whole allocated vCPU on Northflank. Memory is the same story: $0.0212 per GB-hour on Vercel against $0.00833 on Northflank, and both bill it for the whole run.

In the example below, Northflank costs about half of Vercel for an agent that waits, and about a fifth when both CPUs are busy. Runtime's measured CPU at $0.025 a vCPU-hour puts it under both while the agent waits.

Which one lets a sandbox live longest?

A Vercel session runs for up to 24 hours on Pro and 45 minutes on Hobby. The limit applies to each session, so a persistent sandbox that stops and resumes lives on. Northflank states no time limit: a sandbox can run for seconds or weeks. It is created as a service, so it also brings a deployment plan.

A Runtime sandbox has no time limit: it runs while it works. On a paid account, persistent=True keeps it running even while idle, for as long as the account has credit, under a cost ceiling you set. Between tasks it can pause and keep its memory:

Pythonfrom withruntime import Sandboxworker = Sandbox.get_or_create("long-lived-worker", idle_pause_seconds=900)worker.files.write("/workspace/progress.txt", "step 3 of 10\n")print(worker.files.read_text("/workspace/progress.txt"))worker.stop()  # or leave it: it pauses itself after 15 idle minutes

Cost for the same job

The worked case: 1,000 runs on 2 vCPU with 4 GiB (4 GB on both Vercel and Northflank), 60 seconds per run, 20 CPU-seconds of that spent computing.

Vercel$2.12

CPU: 1,000 × 20 s / 3,600 × $0.128; $0.71
Memory: 1,000 × 60 s / 3,600 × 4 × $0.0212; $1.41

Northflank$1.11

CPU: 1,000 × 60 s / 3,600 × 2 × $0.01667; $0.556
Memory: 1,000 × 60 s / 3,600 × 4 × $0.00833; $0.555

Runtime$0.64

CPU: 1,000 × 20 s / 3,600 × $0.025; $0.14
Memory: 1,000 × 60 s / 3,600 × 4 × $0.0075; $0.50

Across 100,000 runs in a month, Vercel totals $212.44, Northflank $111.10 and Runtime $63.89. With both CPUs busy for the whole minute, it is $1.33 on Runtime, $5.68 on Vercel and $1.11 on Northflank. That makes Northflank the cheapest of the three for work that saturates every CPU. Runtime stays under Northflank until a one-minute run keeps its CPUs busy for about 88 of 120 CPU-seconds. Plan fees, creation charges, storage, network and free allowances are excluded.

Which one to pick

Situation	Best fit
The app is on Vercel; up to 10,000 sandboxes on Pro	Vercel Sandbox
CPU-bound jobs that keep every core busy	Northflank
Sandboxes inside your own cloud account	Northflank
Agents that wait on a model	Runtime

In that last case, measured CPU puts Runtime 42% below Northflank and 70% below Vercel. Pause and forks keep memory and running processes.

Moving to Runtime

Vercel Sandbox code, in JavaScript or Python, runs on Runtime once the import changes, without a Vercel project or token:

TypeScriptimport { Sandbox } from "withruntime/vercel"; // was: from "@vercel/sandbox"

Pythonfrom withruntime.vercel import sandbox  # was: from vercel import sandbox

Northflank's service and deployment plan have no counterpart to port. Each sandbox is a single Sandbox.create(). Commands, files and teardown follow map the calls.

A coding agent can do the change from the one instruction in migration, on a branch, tested on the free usage. Credit you add within 30 days of signing up is matched, up to $1,000 in total. Longer reads: Runtime vs Vercel Sandbox, Runtime vs Northflank.

Runtime Cloud

For an agent that mostly waits on a model, Runtime is 70% cheaper than Vercel and 42% cheaper than Northflank. One thousand one-minute runs of a 2 vCPU, 4 GiB sandbox bill $2.12 on Vercel, $1.11 on Northflank and $0.64 on Runtime. Runtime charges measured CPU, as Vercel does, at 80% below Vercel's rate, and its memory rate is the lowest of the three. Rates checked 25 October 2026.

Run npx withruntime login once and the free usage is ready. Work on Vercel Sandbox or Northflank moves with the migration guide.

Sources

Vercel Sandbox pricing and Northflank's sandboxes page rechecked 26 October 2026; the rest checked 25 October 2026.

Facts on this page were checked on 26 October 2026.

Vercel Sandbox vs Northflank sandbox pricing, isolation and limits