# Cirrus Runners

Official pricing: https://cirrus-runners.app/pricing/  
Category: macos · Isolation: apple-vm · DISCONTINUED

## Pricing regimes (raw)


## Features

Yes: idle auto-stop, ≥24 h sessions, nested virtualization, browser, SOC 2, GPU, arm64, macOS, firewall inside (nftables)

No: snapshots, memory snapshots, fork/clone, pause/resume, persistent disk, volumes, custom image (Docker or snapshot), start from your own snapshot, systemd, desktop GUI, computer-use API, code interpreter, browser + desktop control, preinstalled agents, SSH, public IPv4, HTTPS preview URLs, custom domains, raw TCP inbound, private networking, self-hosting, BYOC, open source, Windows, GPU desktop, live fork (no pause), memory fork, automatic snapshots, snapshots on demand, agent harness API, secret proxy, secret proxy for any API, extra volumes, shared volumes

Unknown: everything else. Evidence (source + quote) per feature: https://battleships.dev/data/providers/cirrus-runners.json → feature_evidence

## Caveats

- Discontinued or closed to new customers, confirmed on the live site 2026-09-29. Pricing page unchanged: $150/month per runner, unlimited minutes, 15% annual (12-month) / 50% nonprofit. Home page still says 'Try Cirrus Runners for free' and shows the calculator, but the official Cirrus Labs announcement (cirruslabs.org, 2026-04-07) says no new customers; Cirrus CI shut down 2026
- GitHub Actions runners only; a runner slot runs one job at a time, so 50 concurrent jobs need 50 slots = $7,500/month regardless of hours.
- Engine encodes it as always-on: only priced for always-on instances; for bursty CI the flat fee per concurrency slot is what you pay.
- Low-priority tasks can queue behind the subscription's concurrency. No Windows.
- Cirrus Labs joined OpenAI (announced 2026-04-07); Cirrus Runners no longer accepts new customers, existing customers are served through their contract periods. The pricing page is stale marketing.

## How this provider charges


Cirrus Labs sells GitHub Actions runners as a **flat monthly subscription per concurrent runner with unlimited minutes**. macOS runners are M4 Pro Tart VMs.

## Regime table

| # | Regime | When it applies | How billed | Numbers | Source |
|---|---|---|---|---|---|
| 1 | Monthly per runner | Default | Flat per concurrent runner slot, **no limit on minutes**, change subscription any time | **$150/month** per runner | https://cirrus-runners.app/pricing/ |
| 2 | Annual | 12-month commitment | 15% "Annual Discount" | $127.50/runner/month (derived) | same |
| 3 | Nonprofit | Non-revenue projects | 50% discount | $75/runner/month (derived) | same |
| 4 | Runner types | Same price for each | macOS M4 Pro 4 vCPU/16 GB (paravirtualized GPU); Linux x86 16 vCPU/48 GB (KVM); Linux arm64 8 vCPU/24 GB; Linux GPU 8 vCPU/24 GB + NVIDIA GPU | $150 each | same |
| 5 | Low-priority tasks | Opt-in label | Queue behind normal jobs within your concurrency | $0 extra | same |

## Gotchas

1. You pay for **concurrency, not usage**. 50 parallel jobs need 50 runners even if they run 1 h/day.
2. At 100% utilisation macOS costs about $0.0034/min, the cheapest macOS CI minute anywhere. At low utilisation it is expensive.
3. GitHub Actions only; no Windows.

## Worked example

Workload: 4 vCPU / 8 GiB, 50 concurrent × 8 h/day × 22 days (8,800 h), 30% CPU, 50 GiB snapshots, 100 GiB egress. Illustrative.

- 50 macOS runners (4 vCPU/16 GB): 50 × $150 = **$7,500/month** (annual: $6,375).
- Effective rate: $7,500 / 8,800 h = $0.85/h, or $0.0142/min.
- Snapshots and egress are not metered or published.

Sources: https://cirrus-runners.app/pricing/