CI Bench

Compared · Result set v1 · Measured Oct 7–11, 2026 · Updated October 27, 2026

Best GitHub Actions runners in 2026: 12 options compared (measured)

Maintained by Vitko, Inc., which sells one of the products measured. Nine configurations measured on 12 real CI workloads, ranked by measured speed; three more with published facts only, labelled not measured in v1.

Quick answer

Vitko Runners (warm starts + test splitting) had the highest measured speed (2.3× GitHub-hosted); RunsOn had the lowest cost per job at 100,000 jobs a month (88% lower), and Vitko Runners the lowest among the managed runners. Self-hosted options cost less at high volume and take operations work. Every entry says where another option leads.

Twelve GitHub Actions runner options, ranked by measured speed
RankOptionTypeSpeed vs GitHub-hostedCost per job at 10,000 jobs a monthList $ a minuteStatus
1Vitko Runners (warm starts + test splitting)VitkoManaged2.3x faster84% lower$0.002Measured
2Vitko RunnersVitkoManaged1.7x faster84% lower$0.002Measured
3GitHub-hosted larger runner (4 vCPU)Managed1.6x faster39% higher$0.012Measured
4BlacksmithManaged1.6x faster52% lower$0.004Measured
5Self-hosted runners on Google CloudSelf-hosted1.5x faster59% lowercloud billMeasured
6RunsOnSelf-hosted1.2x faster75% lowern/aMeasured
7Self-hosted runners on AWS (one VM per job)Self-hosted1.2x faster60% lowercloud billMeasured
8UbicloudManaged1.1x faster83% lower$0.00125Measured
9GitHub-hosted runners (ubuntu-latest, 2 vCPU)Managedbaselinebaseline$0.006Measured
–DepotManagednot measured in v1not measured in v1$0.006Not measured in v1
–NamespaceManagednot measured in v1not measured in v1unit minutesNot measured in v1
–WarpBuildManagednot measured in v1not measured in v1$0.004Not measured in v1

How this was done

Speed is the inverse of the time index and cost is the cost index, each the geometric mean over 12 workloads against GitHub-hosted (ubuntu-latest), from results.json. Ranked by speed; cost per job uses list prices dated 2026-10-05. Workflows, method and every raw job record: the repository, method. Prices for the options not measured are copied from each provider's own page on 2026-10-24. We operated the self-hosted options. Maintained by Vitko, Inc., which sells one of the products measured.

The 9 measured options

1. Vitko Runners (warm starts + test splitting) Vitko

Verdict
2.3x faster, 84% lower cost per job: 1st of 9 by speed; 3rd of 9 by cost per job at 100,000 jobs a month. Where others lead: RunsOn cost less per job at 100,000 jobs a month.
Best for
private-repository teams on Ubuntu x64 that want the lowest measured time and cost per job among the managed runners and can add the warm-start and test-splitting steps
Measured
speed 2.3× GitHub-hosted (ubuntu-latest); cost per job 84% lower at 100,000 jobs a month, 84% lower at 10,000, 84% lower at 1,000; lowest median time on cli, fd, guava, monorepo, pydantic, ts, vite, vue; 3 of 360 measured runs failed. All workloads.
Pricing
$0.002 a minute, per second, no per-job minimum; source, read 2026-10-24
Limits
Private beta; Ubuntu 24.04 on x64 only, private repositories only, United States region; no Arm, Windows, macOS or GPU. Warm starts and test splitting need workflow steps. Run by the benchmark maintainer.

2. Vitko Runners Vitko

Verdict
1.7x faster, 84% lower cost per job: 2nd of 9 by speed; 2nd of 9 by cost per job at 100,000 jobs a month. Where others lead: Vitko Runners (warm starts + test splitting) was faster; RunsOn cost less per job at 100,000 jobs a month.
Best for
private-repository teams on Ubuntu x64 that want to change one runs-on line and nothing else
Measured
speed 1.7× GitHub-hosted (ubuntu-latest); cost per job 84% lower at 100,000 jobs a month, 84% lower at 10,000, 84% lower at 1,000; lowest median time on codex-lint; 0 of 360 measured runs failed. All workloads.
Pricing
$0.002 a minute, per second, no per-job minimum; source, read 2026-10-24
Limits
Private beta; Ubuntu 24.04 on x64 only, private repositories only, United States region; no Arm, Windows, macOS or GPU. Run by the benchmark maintainer.

3. GitHub-hosted larger runner (4 vCPU)

Verdict
1.6x faster, 39% higher cost per job: 3rd of 9 by speed; 9th of 9 by cost per job at 100,000 jobs a month. Where others lead: Vitko Runners (warm starts + test splitting) was faster; RunsOn cost less per job at 100,000 jobs a month.
Best for
jobs that need 4 vCPU and 16 GB on GitHub-hosted infrastructure
Measured
speed 1.6× GitHub-hosted (ubuntu-latest); cost per job 39% higher at 100,000 jobs a month, 39% higher at 10,000, 39% higher at 1,000; lowest median time on none of the workloads; 3 of 360 measured runs failed. All workloads.
Pricing
$0.012 a minute, per minute; source, read 2026-10-24
Limits
Higher list price per minute; larger runner, always charged.

4. Blacksmith

Verdict
1.6x faster, 52% lower cost per job: 4th of 9 by speed; 7th of 9 by cost per job at 100,000 jobs a month. Where others lead: Vitko Runners (warm starts + test splitting) was faster; RunsOn cost less per job at 100,000 jobs a month.
Best for
teams that want a managed runner with a low median time on short, bursty and monorepo-style jobs
Measured
speed 1.6× GitHub-hosted (ubuntu-latest); cost per job 52% lower at 100,000 jobs a month, 52% lower at 10,000, 52% lower at 1,000; lowest median time on Burst: 20 jobs at once; 3 of 360 measured runs failed. All workloads.
Pricing
$0.004 a minute, per minute; source, read 2026-10-24
Limits
Measured at 2 vCPU on the configuration and prices listed on its runner page; cost per job at list prices dated 2026-10-05.

5. Self-hosted runners on Google Cloud

Verdict
1.5x faster, 59% lower cost per job: 5th of 9 by speed; 6th of 9 by cost per job at 100,000 jobs a month. Where others lead: Vitko Runners (warm starts + test splitting) was faster; RunsOn cost less per job at 100,000 jobs a month.
Best for
teams that already operate Google Cloud and want pooled VMs under their control
Measured
speed 1.5× GitHub-hosted (ubuntu-latest); cost per job 59% lower at 100,000 jobs a month, 59% lower at 10,000, 59% lower at 1,000; lowest median time on flask; 1 of 360 measured runs failed. All workloads.
Pricing
All-in cloud cost, measured per job (cost rules)
Limits
We operated it for the benchmark; operations time is not in the cost figures. Fixed monthly costs spread over the volume.

6. RunsOn

Verdict
1.2x faster, 88% lower cost per job: 6th of 9 by speed; 1st of 9 by cost per job at 100,000 jobs a month. Where others lead: Vitko Runners (warm starts + test splitting) was faster.
Best for
teams with high job volumes that are happy to run runners in their own AWS account
Measured
speed 1.2× GitHub-hosted (ubuntu-latest); cost per job 88% lower at 100,000 jobs a month, 75% lower at 10,000, 43% higher at 1,000; lowest median time on container; 1 of 360 measured runs failed. All workloads.
Pricing
All-in cloud cost; RunsOn does not meter minutes (source, read 2026-10-24)
Limits
Runs in your AWS account: you operate it and pay the cloud bill; we operated it for the benchmark.

7. Self-hosted runners on AWS (one VM per job)

Verdict
1.2x faster, 62% lower cost per job: 7th of 9 by speed; 5th of 9 by cost per job at 100,000 jobs a month. Where others lead: Vitko Runners (warm starts + test splitting) was faster; RunsOn cost less per job at 100,000 jobs a month.
Best for
teams that already operate AWS and want one VM per job under their control
Measured
speed 1.2× GitHub-hosted (ubuntu-latest); cost per job 62% lower at 100,000 jobs a month, 60% lower at 10,000, 49% lower at 1,000; lowest median time on none of the workloads; 0 of 360 measured runs failed. All workloads.
Pricing
All-in cloud cost, measured per job (cost rules)
Limits
We operated it for the benchmark; operations time is not in the cost figures. Fixed monthly costs spread over the volume.

8. Ubicloud

Verdict
1.1x faster, 83% lower cost per job: 8th of 9 by speed; 4th of 9 by cost per job at 100,000 jobs a month. Where others lead: Vitko Runners (warm starts + test splitting) was faster; RunsOn cost less per job at 100,000 jobs a month.
Best for
teams that want a low per-minute list price from a managed service
Measured
speed 1.1× GitHub-hosted (ubuntu-latest); cost per job 83% lower at 100,000 jobs a month, 83% lower at 10,000, 83% lower at 1,000; lowest median time on none of the workloads; 1 of 360 measured runs failed. All workloads.
Pricing
$0.00125 a minute, per minute (resource use calculated at per-minute granularity, billed monthly); source, read 2026-10-24
Limits
Measured at the standard 2 vCPU tier; a premium tier is listed at a higher price and was not measured.

9. GitHub-hosted runners (ubuntu-latest, 2 vCPU)

Verdict
baseline, baseline cost: 9th of 9 by speed; 8th of 9 by cost per job at 100,000 jobs a month. Where others lead: Vitko Runners (warm starts + test splitting) was faster; RunsOn cost less per job at 100,000 jobs a month.
Best for
no setup: it is the reference every other result is compared with
Measured
speed baseline; cost per job baseline; lowest median time on none of the workloads; 3 of 360 measured runs failed. All workloads.
Pricing
$0.006 a minute, per minute; source, read 2026-10-24
Limits
The baseline for every comparison; included minutes depend on your plan, and the list price applies to private repositories.

Three more, not measured in v1

Depot not measured in v1

Status
Not measured in v2: Depot runners were assigned but did not start jobs for our account during the test window (checks on 2026-10-01 and 2026-10-24).
Pricing
$0.006 a minute; Per second; included: 2,000 GitHub Actions minutes a month on the Developer plan; 20,000 on Startup; fees: Developer plan $20 a month; Startup plan $200 a month. source, read 2026-10-24
Measured numbers
None: not measured in CI Bench v1, so no speed or cost-per-job figure is given. Provider page.

Namespace not measured in v1

Status
Not yet measured; joins in v2.
Pricing
Not stated as a per-minute figure; Unit minutes; included: Team plan: 100,000 unit minutes; Business: 250,000; Developer: pay as you go; 30-day free trial that includes managed GitHub runners; fees: Developer: none (pay as you go); Team $100 a month; Business $250 a month. source, read 2026-10-24
Measured numbers
None: not measured in CI Bench v1, so no speed or cost-per-job figure is given. Provider page.

WarpBuild not measured in v1

Status
Not benchmarked WarpBuild's Terms of Service (§9, Prohibited Uses) prohibit “Use the Service to compete with Argonaut HQ, Inc.” As the maker of a competing runner service, we don't run it. Terms read 2026-10-24.
Pricing
$0.004 a minute; Per minute ("CI runners are billed on a per-minute basis"); included: A free usage or credit amount is mentioned; the amount is not on the pricing page text read; fees: None stated; add-ons priced separately; bring-your-own-cloud runners $0.002 per minute. source, read 2026-10-24
Measured numbers
None: not measured in CI Bench v1, so no speed or cost-per-job figure is given. Provider page.

More: speed ranking · cost ranking · alternatives to GitHub-hosted · self-hosted vs managed · free minutes · list prices · price per minute, 2 and 4 vCPU · Docker builds · CPU-heavy jobs · spot instances · security questions · ARC vs managed · self-hosted platform fee · cut the Actions bill · CLI and MCP · RunsOn vs Blacksmith vs Depot vs Namespace · WarpBuild vs Ubicloud vs BuildJet · alternatives to Depot, WarpBuild, Namespace, BuildJet. Maintained by Vitko, Inc., which sells one of the products measured. Page generated October 27, 2026.