Cheaper GitHub Actions runners 2026 edition
GitHub-hosted Linux runners are billed at $0.008/min for 2 vCPU, $0.016/min for
4 vCPU, and so on - private repos only; public repos are free. Once your team
starts pushing past ~50,000 minutes/month, third-party runner
providers can drop the per-minute cost by 50-90% with a one-line
runs-on: change. This page compares six of them. Pricing was
pulled from each provider's public docs as of April 2026;
always verify on the provider's site before signing.
Affiliate disclosure: some links below pass through
/r/<provider> redirects. If you sign up after clicking,
depmedic may receive a small referral fee at no cost to you. Pricing rows below
are objective and not influenced by referral terms; runner picks are based on
what genuinely cuts your bill the most for the workload type. We do not list
providers we have not personally audited the docs of.
Side-by-side
| Provider | 2-vCPU/min | 4-vCPU/min | vs GH-hosted | OIDC | Cache | Self-host needed? | Notes |
|---|---|---|---|---|---|---|---|
| BuildJet Sign up |
~$0.004 | ~$0.008 | ~50% cheaper | Yes | Faster cache | No - their cloud | Drop-in runs-on: buildjet-2vcpu-ubuntu-2204. AWS-backed. Most mature alt. |
| Namespace Sign up |
~$0.005 | ~$0.010 | ~40% cheaper | Yes | Snapshot cache (very fast) | No - their cloud | Best for cold starts and very fast caching. Used by Linear, Vercel, others. |
| Ubicloud Sign up |
~$0.0008 | ~$0.0016 | ~10x cheaper | Yes | Standard | No - open-source cloud | Lowest sticker price. Open-source control plane. Drop-in runs-on: ubicloud-standard-2. |
| RunsOn Sign up |
your AWS | your AWS | ~10x cheaper | Yes | S3-backed | Runs in your AWS | EC2 Spot in your AWS account; you pay AWS directly. Best for >200k min/mo. |
| WarpBuild Sign up |
~$0.004 | ~$0.008 | ~50% cheaper | Yes | Distributed (very fast) | No - their cloud | Distributed cache layer is the standout. Good Docker layer reuse. |
| Blacksmith Sign up |
~$0.004 | ~$0.008 | ~50% cheaper | Yes | Standard, fast | No - their cloud | Y Combinator-backed. "2x faster, 50% cheaper" is their claim. Drop-in. |
| GitHub-hosted baseline | $0.008 | $0.016 | baseline | Yes | 10 GB free / repo | No | Free for public repos. ubuntu-latest, windows-latest ($0.016+), macos-latest ($0.08+, 10x). |
Quick savings calculator
What would you save?
How to switch (the one-line change)
Most third-party runners are drop-in: change runs-on: in your
workflow file, sign up for the provider, install their GitHub App on your repo
or org, and you're done. No code changes, no Dockerfile, no scripts.
Show example: BuildJet
# before
jobs:
test:
runs-on: ubuntu-latest
# after (BuildJet)
jobs:
test:
runs-on: buildjet-2vcpu-ubuntu-2204
Show example: Ubicloud
jobs:
test:
runs-on: ubicloud-standard-2 # 2 vCPU
# or: ubicloud-standard-4, ubicloud-standard-8, etc.
Show example: Namespace
jobs:
test:
runs-on: namespace-profile-default
# profile defined in your Namespace dashboard
Which one for which workload?
- You just want a one-line cheaper Linux runner: BuildJet or Blacksmith. Both are drop-in, both are mature, both halve the cost.
- You're hitting cold-cache pain: Namespace (snapshot cache) or WarpBuild (distributed cache). Restored caches in 1-3 sec instead of 20+.
- You want the absolute lowest per-minute number: Ubicloud. Trades some maturity for ~10x lower price.
- You're an AWS shop already: RunsOn. Spot instances, you pay AWS directly, pure cost savings, but you take on the ops. Worth it past ~200k min/month.
- You're on macOS-latest because you build iOS apps: stuck on GitHub-hosted or self-hosted Macs - none of the third-party providers above offer Mac runners (yet).
Pair this with the rest of depmedic
- /scan.html - find the workflows in your repo
that are using
macos-latestwhen Linux would do (the single biggest waste source per our daily leaderboard). - /ask.html - paste a workflow, get an AI-suggested conversion to any of the runners above.
- npx gha-budget - estimate the dollar cost of any GHA workflow before you commit it.
- Cut Your CI Bill cookbook ($19) - 30 paste-ready cost optimization patterns.