| Feature | Avrea | GitHub-hosted runners |
|---|---|---|
| Operating systems | Linux, Windows and macOS | Linux, Windows and macOS |
| Workflow files and PR checks | Keep GitHub Actions; change selected runner labels | Use GitHub runner labels |
| Paid Linux x86, 2 vCPU / 8 GB | $0.004/min | $0.006/min |
| Public repositories | Avrea's monthly allowance applies | Standard runners are free |
| Private repositories | 3,000 monthly minutes in 2-vCPU equivalents | Shared compute credit by plan; eligible larger runners included on Team and Enterprise Cloud |
| Caching | Automatic actions/cache integration, build-tool caches and paid persistent disks | GitHub's Actions cache service |
| Debugging | GitHub checks plus Avrea logs, resource metrics and live SSH | Workflow logs and additional debug logging |
| Static IPs | Enterprise feature; $100/IP/month | Linux and Windows larger runners on Enterprise Cloud |
| AI-generated pull requests | Avrea AI proposes CI fixes, tests them and opens PRs | Copilot cloud agent takes on coding tasks and opens PRs |
GitHub’s AI products are separate from its runner offering.
Keep GitHub Actions
After connecting Avrea, change the runs-on label for the jobs you want to move. GitHub continues to handle triggers, dependencies between jobs, pull-request checks and deployment rules. Runner labels and setup.
Our guide to GitHub Actions runners explains hosted, self-hosted and third-party runners. If you are considering a different CI system as well, see the broader GitHub Actions alternatives guide.
Runner pricing
Prices checked 14 October 2026.
Avrea’s paid Linux x86 rates are one-third lower at the sizes below. Its ARM runners have a higher minute rate and the highest measured ARM CPU scores in RunsOn’s 4-vCPU PassMark results. The CPU table shows both architectures.
| Linux runner | Avrea | GitHub-hosted |
|---|---|---|
| x86 · 2 vCPU / 8 GB | $0.004/min | $0.006/min |
| x86 · 4 vCPU / 16 GB | $0.008/min | $0.012/min |
| ARM64 · 2 vCPU / 8 GB | $0.012/min | $0.005/min |
| ARM64 · 4 vCPU / 16 GB | $0.024/min | $0.008/min |
These are paid compute rates before included usage or discounts. The 4-vCPU GitHub rates are for paid larger runners. Avrea pricing, GitHub runner rates, GitHub runner specifications.
Free and included usage
Standard GitHub-hosted runners are free for public repositories, including the standard 4-vCPU / 16-GB x86 and ARM shapes. GitHub plans include $12 to $300 per month for eligible compute. Private and internal repository runners share it with cloud sandbox compute and memory. Eligible larger runners can also draw on it on Team and Enterprise Cloud; larger runners are still charged in public repositories. GitHub Actions billing.
Avrea includes 3,000 monthly minutes in 2-vCPU equivalents and has no monthly base fee. Larger runners consume more of the allowance. Avrea’s allowance is shared with other metered usage, including static IPs and persistent disks. Avrea plan details.
Our GitHub Actions pricing guide covers the other costs, including storage.
Compare cost and feedback time
A ten-minute job at the listed 2-vCPU x86 rates costs $0.04 on Avrea or $0.06 on GitHub before allowances. If it also finishes sooner on Avrea, fewer billed minutes reduce the compute cost further.
Billed execution time excludes queueing. PR feedback also depends on the other jobs required before merging.
CPU benchmarks
Avrea scores higher on both x86 and ARM in RunsOn’s 14 October 2026 PassMark results. These are 4-vCPU / 16-GB runners, with each result the median of five runs.
| 4-vCPU runner | Avrea | GitHub-hosted |
|---|---|---|
| x86 · PassMark single-thread | 4,243 | 2,888 |
| x86 · PassMark CPU Mark, all cores | 11,174 | 7,510 |
| ARM64 · PassMark single-thread | 4,860 | 1,877 |
| ARM64 · PassMark CPU Mark, all cores | 8,388 | 3,974 |
These tests isolate CPU performance. Complete build times also depend on downloads, cache reuse and I/O. RunsOn sells runners and publishes its test code and methodology. RunsOn CPU results. Data: RunsOn, CC BY 4.0; selected rows, shortened labels.
Actions cache benchmarks
In RunsOn’s 14 October 2026 cache results, these 4-vCPU / 16-GB x86 runners saved and immediately restored a 4 GB file:
| Operation | Avrea | GitHub-hosted |
|---|---|---|
| Restore | 14 seconds | 34 seconds |
| Save | 20 seconds | 20 seconds |
Avrea restores faster in this test; save times are equal. This is a cache-transfer benchmark, not an end-to-end build result. RunsOn cache results.
Each result uses five runs. RunsOn sells runners too. Data: RunsOn, CC BY 4.0; selected rows, shortened labels.
Our runner benchmark guide covers setup time, dependency downloads and build-cache reuse alongside these transfer measurements.
Caching
On Avrea, supported actions/cache usage is routed to Avrea’s cache automatically. Package proxies reuse downloads, and integrations for Turborepo, Gradle, Bazel and sccache reuse valid build outputs. That can remove compilation or other tasks from a repeat build, beyond the transfer-time improvement measured above. Avrea cache documentation, build-cache setup.
Existing Actions cache keys and restore logic continue to apply. For Docker builds, Avrea’s persistent disks retain images, layers and builder state between jobs. You can also export layers to its BuildKit cache endpoint. The Docker guide covers both configurations. Persistent disks cost $0.50 per provisioned GB per month, separate from the included repository cache. Docker caching on Avrea.
Our GitHub Actions caching guide covers cache keys, restore behavior and build-tool caches.
File caches and build-output caches
GitHub’s Actions cache restores files selected by the workflow. Turborepo, Gradle and compiler caches can reuse valid build outputs for the current inputs, allowing unchanged tasks to be skipped. GitHub dependency caching.
A large cached directory can take longer to restore than the work it saves. In that case, a narrower cache or a tool-level output cache can reduce job time.
Debugging and AI-assisted fixes
A job can fail without its logs explaining whether it ran out of memory or hit a CPU bottleneck. Avrea’s resource metrics help answer those questions, and SSH lets you inspect the job while it is running. GitHub’s workflow logs and debug logging remain available too.
Avrea AI takes the investigation through to a proposed change: it uses logs, code changes and test history, runs the fix through your full test suite and opens a PR if it passes. Your team reviews the result.
GitHub offers ways to delegate this work too. Copilot cloud agent can take on code changes and open pull requests. GitHub Agentic Workflows can investigate CI failures and suggest fixes, and are now generally available. They require an AI engine and workflow configuration.
Avrea also offers sandboxes for work that needs a remote execution environment.
Avrea offers static IPs at $100 per IP per month, listed under its Enterprise features. GitHub supports static IP ranges for Linux and Windows larger runners on Enterprise Cloud; macOS larger runners do not support them. Avrea static IP pricing, GitHub static IP configuration.
Avrea documents its security controls and GitHub App permissions.
What changed for Zero
Zero’s case study describes moving sixteen GitHub Actions workflow files to Avrea while keeping the existing actions and Docker builds. One PR typecheck fell from five minutes of waiting to one. API deployments became roughly twice as fast without application-code changes.
Related reading
- Avrea pricing: current rates, runner sizes and included features.
- GitHub Actions pricing guide: how to understand and reduce the bill.
- Avrea vs Blacksmith: CPU and cache benchmarks, Docker caching and AI-assisted fixes.
- Avrea vs WarpBuild: equal x86 minute rates, cache benchmarks and BYOC.
- Avrea vs Ubicloud: managed runner prices, standard and premium tiers, caching.
- All runner comparisons: Blacksmith, Depot, Namespace, WarpBuild, Ubicloud and GitHub-hosted runners.
Try one job on Avrea
Follow the setup guide to connect your repository. For a standard 2-vCPU / 8-GB Linux x86 test, change the selected job’s label to avrea-ubuntu-latest. Check the size explicitly if your baseline is a public-repository runner or a paid larger runner; the defaults are not always the same.
Run the same commits several times on both providers with matching resources. Include a cold cache, a warm run and a normal source change. Record the full workflow’s completion time as well as the selected job’s time, since parallel jobs may change what the developer actually waits for.
Try Avrea · Runner sizes and pricing