GitLab Runner vs GitLab-hosted runners
GitLab Runner suits teams on GitLab CI/CD that want to run jobs on their own machines, clusters or autoscaled cloud instances. GitLab-hosted runners suits teams on GitLab.com that want CI jobs to run with no runner setup and can work within the included compute minutes. GitLab Runner is an open-source project you run yourself for GitLab CI; GitLab-hosted runners is the CI platform’s own hosted runners for GitLab CI, priced at $10 per 1,000 compute minutes.
Key differences between GitLab Runner and GitLab-hosted runners
- Type: GitLab Runner is an open-source project you run yourself; GitLab-hosted runners is the CI platform’s own hosted runners.
- Isolation: GitLab Runner: Depends on executor. GitLab-hosted runners: Ephemeral VM.
- Open source: GitLab Runner is open source (MIT); GitLab-hosted runners is not.
- Self-hosting: GitLab Runner documents a self-hosted option; GitLab-hosted runners does not.
- Your own cloud account: GitLab Runner documents running in your own cloud account; GitLab-hosted runners does not.
What they have in common
- Both are from GitLab Inc..
- Both run GitLab CI jobs.
- Both document Linux, macOS and Windows runners.
- Both document arm64 runners.
- Both document GPU runners.
GitLab Runner vs GitLab-hosted runners at a glance
| Attribute | GitLab Runner | GitLab-hosted runners |
|---|---|---|
| What it is | The open-source agent that runs GitLab CI/CD jobs, with executors for Docker, Kubernetes, shell and autoscaled cloud instances. | GitLab.com’s built-in runners: each job gets a single-use VM, metered in compute minutes with a cost factor per machine size. |
| Type | Open-source project | First-party hosted runners |
| CI systems | GitLab CI | GitLab CI |
| Operating systems | Linux, macOS, Windows | Linux, macOS, Windows |
| Architectures | x64, arm64 | x64, arm64 |
| Isolation | Depends on the executor: Docker and Kubernetes run jobs in containers, the Docker Autoscaler and Instance executors create cloud instances through fleeting plugins, and the Shell and SSH executors run on the host without a clean build environment | Each job runs in a newly provisioned VM that serves a single job and is deleted right after it; firewall rules block inbound traffic and traffic between VMs |
| Fresh environment per job | Not documented | Yes |
| Hardware | Not documented | Linux runners are Google Compute Engine VMs running Container-Optimized OS; macOS runners are Apple silicon M1 and M2 Pro machines. GitLab notes that machine and processor types might change |
| Performance (vendor claim) | Not documented | Not documented |
| Runner sizes | Not documented | Linux x86-64: 2 vCPU / 8 GB to 32 vCPU / 128 GB; Linux Arm64: 2 vCPU / 8 GB to 8 vCPU / 32 GB; macOS: 4 vCPU / 8 GB (M1) and 6 vCPU / 16 GB (M2 Pro); Windows: 2 vCPU / 7.5 GB |
| Caching | The docs describe a distributed cache that uses an object storage server (the example uses S3) to share the cache between Docker hosts, and a Docker registry mirror for autoscaled hosts. | Hosted runners share a distributed cache in Google Cloud Storage. Cache contents not updated in the last 14 days are removed, and an uploaded cache archive can be at most 5 GB. |
| Docker layer caching | Not documented | Not documented |
| GPU | GPU use is documented for the Shell, Docker (gpus option with the NVIDIA Container Toolkit), Kubernetes and Docker Machine executors. | One GPU machine type on Linux x86-64 for Premium and Ultimate: 4 vCPU, 15 GB RAM and 1 NVIDIA Tesla T4 (or similar) with 16 GB, at cost factor 7. |
| Concurrency | Set by the concurrent setting in config.toml; for autoscaling the docs set it to max_instances multiplied by capacity_per_instance | Maximum jobs in active pipelines on GitLab.com: 500 on Free, 20,000 on Premium and 60,000 on Ultimate |
| Static IPs | Not documented | Not documented |
| SSH into running jobs | Not documented | Not documented |
| Pricing | Free and open source; you pay for the infrastructure it runs on. The docs list it for the Free, Premium and Ultimate tiers. | No per-minute price is published. Jobs consume compute minutes at job duration times a cost factor: Linux small 1, medium 2, large 3, xlarge 6, 2xlarge 12; GPU 7; macOS M1 6 and M2 Pro 12 (beta); Windows 1 (beta). Additional compute minutes cost $10 per 1,000 minutes as a one-time purchase. |
| Free tier | Not documented | 400 compute minutes per month on the Free tier; Premium includes 10,000 and Ultimate 50,000 per month |
| Open source | Yes (MIT) | No |
| Self-host / your cloud | Yes / Yes | No / No |
| Compliance | Not documented | Not documented |
| Regions | Not documented | Linux x86-64 VMs run in Google Cloud us-east1 and us-east5; Linux Arm64 VMs run in us-central1 |
| Company headquarters | Not documented | Not documented |
| Good fit for | Teams on GitLab CI/CD that want to run jobs on their own machines, clusters or autoscaled cloud instances | Teams on GitLab.com that want CI jobs to run with no runner setup and can work within the included compute minutes |
Verified November 22, 2026 from each provider’s documentation. Performance figures are vendor claims, not our measurements.
Which should you choose?
Reasons to choose GitLab Runner
- You use GitLab CI/CD and want to run jobs on infrastructure you control
- You want a choice of executors, from containers on Docker or Kubernetes to full cloud instances
- You want autoscaling on AWS, Google Cloud or Azure with plugins that GitLab maintains
- You need runners on Linux, macOS and Windows from the same project
Reasons to choose GitLab-hosted runners
- You host code on GitLab.com and want pipelines to run without registering any runner
- You want each job in its own VM that is deleted after the job
- You need Docker-in-Docker, which the Linux runners support by running in privileged mode
- You need Linux Arm64 or a GPU machine alongside x86-64 from the same platform
Trade-offs to check first
- GitLab Runner: The Docker Machine executor is deprecated and scheduled for removal in GitLab 20.0
- GitLab Runner: SSH, Shell, Parallels, VirtualBox, Custom and Docker Machine executors receive critical security updates only, with no new features planned
- GitLab-hosted runners: Hosted runners on macOS and on Windows are in beta
- GitLab-hosted runners: Jobs time out after 3 hours regardless of the project timeout
Run your slowest workflow on both before deciding. See the full GitLab Runner profile and GitLab-hosted runners profile, or the complete comparison.
More comparisons
- Elastic CI Stack for AWS vs GitLab Runner
- GitLab Runner vs terraform-aws-gitlab-runner
- AWS CodeBuild-hosted runners vs GitLab-hosted runners
- Buildkite hosted agents vs GitLab-hosted runners
- GitHub-hosted runners vs GitLab-hosted runners
- GitLab-hosted runners vs Microsoft-hosted agents
- GitLab-hosted runners vs RunJob
Frequently asked questions
Is GitLab Runner or GitLab-hosted runners better?
What is the difference between GitLab Runner and GitLab-hosted runners?
Which is open source, GitLab Runner or GitLab-hosted runners?
How do GitLab Runner and GitLab-hosted runners isolate jobs?
How does GitLab Runner pricing compare with GitLab-hosted runners?
Do GitLab Runner and GitLab-hosted runners offer GPU runners?
Where do GitLab Runner and GitLab-hosted runners runners run?
Sources
- docs.gitlab.com/runner/
- docs.gitlab.com/runner/executors/
- docs.gitlab.com/runner/executors/docker_autoscaler/
- docs.gitlab.com/runner/executors/instance/
- docs.gitlab.com/ci/runners/hosted_runners/
- docs.gitlab.com/ci/runners/hosted_runners/linux/
- docs.gitlab.com/ci/runners/hosted_runners/macos/
- docs.gitlab.com/ci/runners/hosted_runners/windows/
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). GitLab Runner vs GitLab-hosted runners. cirunners.org. Last reviewed November 22, 2026. https://cirunners.org/compare/gitlab-runner-vs-gitlab-hosted-runners/