Runtime

Vercel Sandbox alternatives: agent sandboxes ranked by cost

Runtime is the cheapest Vercel Sandbox alternative, 70% less per job, and it needs no Vercel team or project.

Runtime bills active CPU like Vercel Sandbox, at about a fifth of the price. 1,000 one-minute runs of a 2 vCPU, 4 GiB sandbox cost $0.64 on Runtime and $2.12 on Vercel, $149 a month saved at 100,000 runs. Both run Firecracker microVMs, and Vercel Sandbox code keeps working after one import change. Rates checked 23 September 2026.

Estimated compute cost

70% less than Vercel Sandbox53% less than Cloudflare Sandbox77% less than E2B

2 vCPUs · 4 GiB · 60 s a run · 20 CPU-seconds busy

1,000

Compute only, at list rates checked 23 September 2026 (Vercel Sandbox pricing, Cloudflare Containers pricing, E2B pricing). Plan fees, network, taxes and free credits are left out. The worked example below shows the arithmetic.

Why do teams look beyond Vercel Sandbox?

Vercel Sandbox already bills active CPU rather than every vCPU it holds, so the reasons to move are the rate and what a sandbox keeps:

  • The CPU rate. Active CPU is $0.128 an hour in iad1, against $0.025 on Runtime. The gap grows as work gets busier: with both CPUs busy for a whole minute, 1,000 runs cost $5.68 on Vercel.
  • Memory is billed in whole minutes. Vercel bills provisioned memory in 1-minute minimum increments, at $0.0212 per GB-hour, and each vCPU comes with 2 GB.
  • A stop loses the processes. A stopped Vercel sandbox keeps its filesystem, and a snapshot or fork captures the filesystem, so processes start again. A paused Runtime sandbox wakes with its processes still running.
  • Ports are public. Vercel's exposed ports answer at a public URL. A Runtime preview needs a token unless you make it public.
  • It needs a Vercel team. Usage beyond the Hobby allowance needs Pro, and a Hobby session lasts 45 minutes. Once the Hobby allowance is used up, sandbox creation is paused until the next cycle.

Which alternatives are cheapest?

Here the job is priced at each provider's list rates: 1,000 runs of 60 seconds on a 2 vCPU, 4 GiB sandbox, with 20 CPU-seconds of work in each. Vercel lands sixth, ahead of every provider that bills held vCPUs, because the agent is mostly waiting. Each total is worked out on pricing, without plan fees, storage, network, credits or custom deals.

# Provider Isolation CPU billed on 1,000 runs
1 Runtime Firecracker microVM, own kernel Measured CPU, with a small floor $0.64
2 Northflank microVMs for CPU work, gVisor for GPU work Allocated vCPUs $1.11
3 Cloudflare Sandbox (2 vCPU, 6 GiB, 12 GB disk) A container in its own VM Active CPU $1.35
4 Fly Machines, performance-2x with 4 GB Firecracker microVM The machine's size while started $1.44
5 Morph, 2 MCUs Full virtual machines The machine's size in MCUs $1.67
6 Vercel Sandbox, iad1 (4 GB) (the one you are replacing) Firecracker microVM, own kernel Active CPU $2.12
7 CodeSandbox SDK, a Nano VM Firecracker microVM The VM's size, by the minute rounded up $2.48
8 E2B Firecracker microVM, own kernel Allocated vCPUs $2.76
9 Daytona Containers by default; VM class separate Allocated vCPUs $2.76
10 Blaxel (4 GB) Lightweight virtual machines Included with memory, active time $2.76
11 Fly Sprites (4 GB in use) Firecracker microVM Measured CPU $3.31
12 Modal Sandboxes (1 physical core) gVisor, a user-space kernel Requested or used, whichever is higher $3.97
13 Runloop, a MEDIUM devbox microVM, with a container inside Allocated CPUs while running $5.33

Cost for the same job

The working for Runtime, Vercel and the two other providers that bill active CPU or run a microVM per sandbox:

TextRuntime     CPU    1,000 × 20 s / 3,600 × $0.025        = $0.14            Memory 1,000 × 60 s / 3,600 × 4 × $0.0075   = $0.50            Total                                          $0.64Vercel      CPU    1,000 × 20 s / 3,600 × $0.128        = $0.71            Memory 1,000 × 60 s / 3,600 × 4 × $0.0212   = $1.41            Total                                          $2.12Cloudflare  CPU    1,000 × 20 s × $0.000020             = $0.40            Memory 1,000 × 60 s × 6 × $0.0000025        = $0.90            Disk   1,000 × 60 s × 12 × $0.00000007      = $0.05            Total                                          $1.35E2B         CPU    1,000 × 60 s / 3,600 × 2 × $0.0504   = $1.68            Memory 1,000 × 60 s / 3,600 × 4 × $0.0162   = $1.08            Total                                          $2.76

Vercel's CPU rate is what bites as the work gets heavier: busy for the whole minute on both CPUs, the job costs $1.33 on Runtime, $5.68 on Vercel, $3.35 on Cloudflare and $2.76 on E2B, and Vercel drops from third of the four to last. A month of 100,000 waiting runs is $63.89 on Runtime and $212.44 on Vercel.

What each alternative changes for a Vercel user

1. Runtime: the same microVM, a fifth of the CPU rate

Runtime also bills CPU by use, at $0.025 per vCPU-hour against Vercel's $0.128, with a floor of a twentieth of a vCPU, and reserved memory at $0.0075 per GiB-hour. After price, the biggest change is what a pause keeps: files, memory and the running processes, for 1 to 365 days, and a fork copies all three. No hosting plan, team or project sits in front of it; the account is prepaid credit from $10.

2. Northflank: under Vercel however hard the CPUs work

Its allocation rates, $0.01667 per vCPU-hour and $0.00833 per GB-hour, make this job $1.11 whether the agent idles or computes. Sandboxes are services with deployment plans, and they can be deployed into your AWS, GCP or Azure account.

3. Cloudflare Sandbox: the other edge platform

Cloudflare also meters active CPU, at $0.072 per vCPU-hour, just over half of Vercel's rate. It insists on 3 GiB of memory per vCPU, runs from a Worker on the Workers Paid plan and forgets a sandbox's files after 10 idle minutes unless they were copied to R2.

4. Fly Machines: a bill that ignores CPU load

Where Vercel's charge rises with CPU, a Machine's stays flat: performance-2x with 4 GB is $0.0861 an hour in iad, so $1.44 busy against Vercel's $5.68. Its 18 regions come close to Vercel's 19; the root disk is temporary.

5. Morph: forks that carry memory

A Vercel fork copies the filesystem. Morph's Infinibranch copies a running machine together with its memory, into hundreds of replicas. It charges $0.05 per MCU-hour, and this job takes 2.

6. CodeSandbox SDK: sized VMs, capped concurrency

A Nano VM of 2 cores and 4 GB costs $0.1486 an hour, billed per minute and rounded up. For a Vercel Pro user used to 10,000 at once, the free plan's 10 VMs is the gap; Scale, at $170 a month, raises it.

7. E2B: dearer while waiting, cheaper when busy

Allocation billing at $0.0504 per vCPU-hour and $0.0162 per GiB-hour gives $2.76 in both cases, above Vercel's waiting figure and below its busy one. Hobby sessions end after an hour, and the Apache-2.0 runtime can be self-hosted.

8. Daytona: GPUs, Windows and EU regions

E2B's rates, a container unless you ask for its VM class, and a wider range of machines: GPU sandboxes, Windows machines, US and EU regions, and $200 of free compute.

9. Blaxel: no CPU meter at all

Active memory costs $0.0000115 per GB-second and brings its cores with it, 4 for 8 GB. Standby follows the last closed connection, and waking takes about 25 ms.

10. Fly Sprites: active CPU, memory in use

Fly.io's agent sandboxes meter CPU much as Vercel does, at $0.07 per CPU-hour, and memory in use at $0.04375 per GB-hour. A Sprite pauses by itself, keeps its memory at first and keeps its 100 GB disk for good.

11. Modal Sandboxes: one account for GPUs and functions

gVisor in place of a microVM, and CPU charged on the greater of request and use at $0.1419 per physical core-hour. The draw is GPUs, functions and scheduled jobs under one SDK.

12. Runloop: devboxes with benchmarks ready

A container inside a microVM, at $0.108 per CPU-hour and $0.0252 per GB-hour plus disk, with SWE-Bench and other public benchmarks set up to run.

How to switch from Vercel Sandbox

Runtime's SDK runs code written for Vercel Sandbox's SDK, in JavaScript and Python: commands, files, ports, snapshots and persistent sandboxes, with no Vercel project or token. Change the import and set RUNTIME_API_KEY, or run npx withruntime login once:

TypeScriptimport { Sandbox } from "withruntime/vercel"; // was: from "@vercel/sandbox"
Pythonfrom withruntime.vercel import sandbox  # was: from vercel import sandbox

Sandboxes keep Vercel's defaults (2 vCPUs, 2048 MiB each, a 5-minute timeout) and persistence, so stop() pauses one and Sandbox.get({ name }) wakes it. Drives, other regions or a timeout over an hour throw NotSupportedError before anything happens. The full mapping is in the Vercel comparison.

Ports that must stay private can move to a Runtime preview, which asks for a token by default:

TypeScriptimport { Sandbox } from "withruntime";await using sbx = await Sandbox.create();await sbx.spawn("python3 -m http.server 3000");const preview = await sbx.previews.create(3000);const page = await fetch(preview.url, {  headers: { "x-runtime-preview-token": preview.token! },});console.log(page.status);

Prefer to delegate? The instruction in migration tells a coding agent to move the project on a branch, check it on the free trial and report what a month would cost. Run npx withruntime switch --from vercel before adding credit the first time, and that first amount is matched, up to $100. Head-to-heads: E2B vs Vercel Sandbox, Daytona vs Vercel Sandbox and Modal vs Vercel Sandbox.

Sources

Every rival rate below was checked 23 September 2026, and Vercel's own pricing page again on 25 September 2026.

Claim 50 hours free