Cirrus Runners vs compiler.dev: Flat or Metered

Last updated 22 October 2026

Cirrus Runners charges a flat $150 a month for each concurrent runner, with unlimited minutes. compiler.dev charges by the minute. A flat price wins when a runner is busy most of the day, and per-minute billing wins when it is not. The break-even for a Linux machine is a few thousand minutes a month, so the first job is to count your minutes. Prices were checked in October 2026 on Cirrus Runners' pricing page.

What Cirrus Runners is

Cirrus Runners is a hosted runner service from Cirrus Labs. Each runner is a single-use virtual machine destroyed after each job. Four types are sold at the same $150 a month: macOS with 4 M4 Pro cores and 16 GB, Linux x86 with 16 vCPU and 48 GB, Linux arm64 with 8 vCPU and 24 GB, and a GPU Linux runner with 8 vCPU and 24 GB. The company builds its own macOS (Tart) and Linux (Vetu) virtualization. The pricing page claims 2 to 3 times the speed of GitHub's managed runners, which is a vendor claim we have not verified.

One runner is one concurrent job. To run five jobs in parallel you buy five. Sizes inside a type use a concurrency weight: for example the medium Linux x86 label counts as 0.5, the small Arm label as 0.25, and macOS XL (8 CPUs and 32 GB) as 2.0. There is a 10-day free trial after you install the app, and discounts of 15% for annual commitment or 50% for non-profits that earn no revenue.

A banner on Cirrus Labs' site at the time of checking says the company is joining OpenAI to work on agent infrastructure. If you are choosing a long-term vendor, read that notice and ask what it means for the runner service.

Side-by-side

	Cirrus Runners	compiler.dev
Price model	$150 per runner per month, unlimited minutes	Per minute, no monthly fee
Linux x64	16 vCPU / 48 GB per runner	2/4/8/16/32 vCPU at $0.003/$0.006/$0.012/$0.024/$0.048 per min (8 vCPU is 16 GB)
ARM Linux	8 vCPU / 24 GB	Not offered
macOS	4 M4 Pro cores / 16 GB, same $150	M4 Pro 14-core $0.055/min, 7-core $0.028/min (early access)
GPU	Nvidia, 8 vCPU / 24 GB	Not offered
Per-minute rate	Null (flat)	$0.012 at 8 vCPU
Caching	10 GB included per runner, shared; Omni Cache; Docker layer cache via a local endpoint	Dependency, git and Docker caches; Nx and Turborepo remote cache
Free tier	10-day trial	200 minute trial, card required
Setup	Install app, Stripe checkout, change label	Install app, change one line
Fallback	Not described	Original runners

The break-even math

A month has about 43,200 minutes. One Cirrus Linux x86 runner at $150 gives you as much of that as you can use. The comparable compiler.dev size is 16 vCPU at $0.024 a minute.

$150 divided by $0.024 is about 6,250 minutes, or roughly 104 hours. Below that, compiler.dev costs less.
At 8 vCPU ($0.012), the break-even is about 12,500 minutes, or 208 hours, but the Cirrus machine is bigger, so the 16 vCPU line is the fairer one.
Running a compiler.dev 16 vCPU machine around the clock for a month would cost about $778, which is why flat pricing is attractive for always-busy runners.

The catch is utilisation. A single flat runner handles one job at a time. If your team opens ten pull requests at once, they queue behind one runner or you buy ten and pay $1,500 a month. compiler.dev scales with the minutes you use. Count your peak concurrency, not just your average, before you buy flat. The CI minutes tools and the cost calculator help with the monthly count.

Where Cirrus Runners is the better choice

Heavy macOS. One M4 Pro runner for $150 a month with unlimited minutes is much cheaper than any per-minute macOS rate once a runner is busy. At $0.055 a minute on compiler.dev, a month of continuous use is roughly $2,376; at $0.08 on other providers it is far more. Mobile teams with long Xcode builds gain most.
Predictable bills. A fixed charge per runner, no overage surprises.
A busy, small queue. Two or three always-on runners with steady load.
ARM or GPU. Both are offered at the same flat price.
Docker and Gradle caching. Its docs describe layer caching through a local endpoint and a Gradle remote cache.

Where compiler.dev is the better choice

Bursty or low usage. Pay for what runs, and prepaid credit does not expire.
Many parallel jobs. Concurrency is not tied to a count of runners you bought.
A fallback. If compiler.dev cannot run a job or credit ends, your original runners take it.
Different Linux sizes. Five sizes from 2 to 32 vCPU, so a lint job does not pay for a 16 vCPU machine.
Full-pipeline measurements. Our benchmarks show median 1.5× faster across 18 measured stacks (15 faster, 2 the same, 1 slower) pipelines than GitHub-hosted. We did not benchmark Cirrus, and its page claims its own gains; we cannot compare them.

How to switch

Cirrus labels are image names, per its setup page:

jobs:
  test:
    runs-on: ghcr.io/cirruslabs/ubuntu-runner-amd64:24.04-md
  ios:
    runs-on: ghcr.io/cirruslabs/macos-runner:tahoe

To compiler.dev:

jobs:
  test:
    runs-on: [self-hosted, Linux, compiler-ci]

Install the app at app.compiler.dev. Your macOS jobs can stay on Cirrus while Linux jobs move, because runs-on is set per job. If you use Cirrus's Omni Cache sidecar or its local Docker cache endpoint, replace those steps with standard actions before you move. Docker layer caching in CI lists the usual options. See the full provider ranking too.

FAQ

How much does Cirrus Runners cost?

$150 a month per runner, with unlimited minutes (October 2026). A runner is one concurrent job. There is a 15% annual discount and a 50% non-profit discount, which cannot be combined.

Is Cirrus Runners cheaper than compiler.dev?

For a runner that is busy more than about 104 hours a month on a 16 vCPU size, yes. For lower or bursty use, per-minute billing is cheaper.

Does Cirrus Runners have a free trial?

Yes, 10 days after installing the app, according to its setup page.

Does compiler.dev have flat pricing?

No. It bills per minute from prepaid credit. A dedicated M4 Pro host is available from $3,595 a month for macOS, which is a different, host-based model.