AWS CodeBuild-hosted runners vs RunsOn
AWS CodeBuild-hosted runners suits teams on AWS that want GitHub Actions, GitLab or Buildkite jobs to run inside their own account with IAM and VPC access. RunsOn suits teams on GitHub Actions that want ephemeral EC2 runners, caches and networking kept inside their own AWS account under a flat annual licence. AWS CodeBuild-hosted runners is a cloud provider service that runs in your cloud account for GitHub Actions, GitLab CI and Buildkite, priced at $0.005/min; RunsOn is a product that runs runners in your own cloud account or on your own hardware for GitHub Actions, priced at €300/yr licence + AWS cost.
Key differences between AWS CodeBuild-hosted runners and RunsOn
- Type: AWS CodeBuild-hosted runners is a cloud provider service that runs in your cloud account; RunsOn is a product that runs runners in your own cloud account or on your own hardware.
- CI systems: Only AWS CodeBuild-hosted runners documents GitLab CI and Buildkite.
- Isolation: AWS CodeBuild-hosted runners: Fresh build environment. RunsOn: EC2 instance.
- Published price: $0.005/min (AWS CodeBuild-hosted runners) versus €300/yr licence + AWS cost (RunsOn).
- Largest runner: 72 vCPU / 144 GB (AWS CodeBuild-hosted runners) versus 896 vCPU (RunsOn).
- Self-hosting: RunsOn documents a self-hosted option; AWS CodeBuild-hosted runners does not.
- Company headquarters: Amazon Web Services, Inc. is based in Seattle, United States; RunsOn is based in Thorigné-Fouillard, France.
What they have in common
- Both run GitHub Actions jobs.
- Both document Linux and Windows runners.
- Both document arm64 runners.
- Both give every job a fresh environment.
- Neither is open source.
- Both document Docker layer caching.
- Both document GPU runners.
AWS CodeBuild-hosted runners vs RunsOn at a glance
| Attribute | AWS CodeBuild-hosted runners | RunsOn |
|---|---|---|
| What it is | AWS CodeBuild projects that start a single-use runner in your AWS account for each GitHub Actions, GitLab or Buildkite job. | Self-hosted GitHub Actions runner control plane installed in your own AWS account, launching one fresh EC2 instance per job. |
| Type | Cloud provider service | Runs in your cloud or on your hardware |
| CI systems | GitHub Actions, GitLab CI, Buildkite | GitHub Actions |
| Operating systems | Linux, Windows | Linux, Windows |
| Architectures | x64, arm64 | x64, arm64 |
| Isolation | Each job starts a CodeBuild build in a fresh environment that is discarded when it finishes; runner builds use CodeBuild containers or instances on Amazon EC2, or AWS Lambda compute | Each job runs on an EC2 instance that no other job has used and that is terminated when the job ends |
| Fresh environment per job | Yes | Yes |
| Hardware | On-demand Amazon EC2 or AWS Lambda compute, or reserved capacity fleets, on x86-64 and Arm; AWS says runners can use its latest instance types, including ARM-based instances | Any EC2 instance family selected by label, including Graviton and families with local NVMe; the docs cite 1 to 896 vCPUs. |
| Performance (vendor claim) | Not documented | RunsOn says Rust builds run 2.0x, TypeScript 2.1x and Docker builds 2.6x faster than on GitHub-hosted runners in its own benchmarks (vendor claim) |
| Runner sizes | On-demand EC2: 2 vCPU / 4 GiB to 72 vCPU / 144 GiB on Linux x86-64, 2 vCPU / 4 GiB to 48 vCPU / 96 GiB on Arm, 4 vCPU / 8 GiB to 72 vCPU / 144 GiB on Windows. Lambda: 1 to 10 GiB of memory | Predefined runners from 1 to 64 vCPU on x64 and arm64; labels can request any EC2 instance type, which the docs describe as 1 to 896 vCPUs |
| Caching | CodeBuild projects can cache to Amazon S3 or use a local cache with source, Docker layer and custom modes. A Docker image server with a shared layer cache is billed separately per second. | Magic Cache redirects actions/cache and Buildx type=gha to an S3 bucket in your account with no size cap (extras=s3-cache). Docker layers can also be cached in a per-stack ECR registry or on sticky EBS disks, and v3.6.0 added S3-backed remote caches for Bazel, Go, Gradle, sccache and Turborepo. |
| Docker layer caching | Yes | Yes |
| GPU | Linux GPU compute types gpu_small (4 vCPU, 16 GiB) at $0.05 and gpu_large (32 vCPU, 255 GiB) at $0.30 per build minute in US East (N. Virginia); GPU models are not stated on the pages checked. | GPU images with NVIDIA drivers are published for x64 and Graviton instances (examples use g4dn, g5 and g5g); GPU compute is billed by AWS and no RunsOn GPU price is published. |
| Concurrency | Concurrently running builds are capped by per-Region service quotas for each environment type, which AWS lists as adjustable | No RunsOn concurrency limit is documented; the Flex mode is recommended for up to about 50,000 jobs a day |
| Static IPs | Not documented | Yes |
| SSH into running jobs | Not documented | Yes |
| Pricing | Standard CodeBuild rates per build minute in US East (N. Virginia): Linux x86-64 small (2 vCPU) $0.005, medium $0.01, large $0.02; Arm small $0.0034, medium $0.007, large $0.015; Windows medium $0.018, large $0.036. EC2 builds are rounded up to the minute; Lambda compute is billed per second. Rates vary by Region. | Flat annual licence per legal entity: Starter €300/year (under 50k jobs/month), Growth €900/year (under 200k), Scale €1,800/year (under 500k), Sponsorship €3,600/year (unlimited jobs, source access). EC2, EBS and S3 are billed by AWS with no RunsOn markup; RunsOn says the control plane costs from about $7/month. |
| Free tier | 100 build minutes per month on general1.small or arm1.small on-demand EC2, or 6,000 build seconds per month on 1 GB Lambda compute | Free licence for non-commercial use (open source, non-profit, educational, personal) with a public link back; 15-day trial on Starter, Growth and Scale |
| Open source | No | No |
| Self-host / your cloud | No / Yes | Yes / Yes |
| Compliance | Not documented | Not documented |
| Regions | All AWS Regions where CodeBuild is available | Not documented |
| Company headquarters | Seattle, Washington, United States | Thorigné-Fouillard, Brittany, France |
| Good fit for | Teams on AWS that want GitHub Actions, GitLab or Buildkite jobs to run inside their own account with IAM and VPC access | Teams on GitHub Actions that want ephemeral EC2 runners, caches and networking kept inside their own AWS account under a flat annual licence |
Verified November 22, 2026 from each provider’s documentation. Performance figures are vendor claims, not our measurements.
Which should you choose?
Reasons to choose AWS CodeBuild-hosted runners
- You want CI jobs to run inside your own AWS account with IAM roles and VPC access
- You use GitHub Actions, GitLab or Buildkite and want one runner service for all of them
- You want single-use runners without operating an autoscaler
- You need Arm, GPU or Windows compute billed at standard CodeBuild rates
Reasons to choose RunsOn
- You want every job on a fresh EC2 instance inside your own AWS account, with code, secrets, caches and logs staying there
- You prefer a flat annual licence by job volume over per-minute fees, with compute billed directly by AWS
- You need Linux x64, arm64, Windows or GPU runners chosen per job through labels
- You need static egress IPs, private subnets, or SSM and SSH access to running jobs
Trade-offs to check first
- AWS CodeBuild-hosted runners: macOS is not listed among the supported runner platforms, and on-demand fleets do not support macOS
- AWS CodeBuild-hosted runners: Build duration on EC2 compute is rounded up to the nearest minute
- RunsOn: AWS only, and macOS is not supported: macOS jobs stay on GitHub-hosted runners
- RunsOn: The server and agent are proprietary; full source is offered only with the €3,600/year Sponsorship licence
Run your slowest workflow on both before deciding. See the full AWS CodeBuild-hosted runners profile and RunsOn profile, or the complete comparison.
More comparisons
- Actions Runner Controller (ARC) vs AWS CodeBuild-hosted runners
- AWS CodeBuild-hosted runners vs Elastic CI Stack for AWS
- AWS CodeBuild-hosted runners vs GitLab-hosted runners
- AWS CodeBuild-hosted runners vs Managed DevOps Pools
- AWS CodeBuild-hosted runners vs terraform-aws-github-runner
- Actions Runner Controller (ARC) vs RunsOn
- Actuated vs RunsOn
- Cirun vs RunsOn
- HyperEnv for GitHub Actions Runner vs RunsOn
- RunsOn vs Sprinters
- RunsOn vs terraform-aws-github-runner
- RunsOn vs WarpBuild
Frequently asked questions
Is AWS CodeBuild-hosted runners or RunsOn better?
What is the difference between AWS CodeBuild-hosted runners and RunsOn?
Which is open source, AWS CodeBuild-hosted runners or RunsOn?
How do AWS CodeBuild-hosted runners and RunsOn isolate jobs?
How does AWS CodeBuild-hosted runners pricing compare with RunsOn?
Is AWS CodeBuild-hosted runners or RunsOn faster?
Do AWS CodeBuild-hosted runners and RunsOn cache Docker layers?
Do AWS CodeBuild-hosted runners and RunsOn offer GPU runners?
Where do AWS CodeBuild-hosted runners and RunsOn runners run?
Sources
- docs.aws.amazon.com/codebuild/latest/userguide/runners.html
- docs.aws.amazon.com/codebuild/latest/userguide/action-runner-overview.html
- docs.aws.amazon.com/codebuild/latest/userguide/action-runner-questions.html
- docs.aws.amazon.com/codebuild/latest/userguide/sample-github-action-runners-update-labels.html
- runs-on.com/
- runs-on.com/pricing/
- runs-on.com/docs/
- runs-on.com/docs/flex-vs-fleet/
Check before you rely on it. Products change quickly, and this page reflects public documentation as of November 22, 2026. Confirm anything that matters in the provider’s own documentation.
Cite this page
CI Runners. (2026). AWS CodeBuild-hosted runners vs RunsOn. cirunners.org. Last reviewed November 22, 2026. https://cirunners.org/compare/aws-codebuild-runners-vs-runs-on/