CI Bench

Workload ranking · Result set v1 · Oct 7–11, 2026 · Updated October 28, 2026

A burst of 20 GitHub Actions jobs: queue time and cost

What happens when many jobs start together, on nine runner configurations.

Short answer

In the CI Bench v1 burst workload (20 jobs dispatched at once), the lowest median time was Blacksmith at 97 s; GitHub-hosted (ubuntu-latest) took 259 s. Queue time varied widely: median 3.0 s on GitHub-hosted against 18.0 s on Vitko Runners, whose p95 queue was 294.6 s against GitHub-hosted's 31.3 s. One burst of 20 on default capacity, so treat it as one data point. Maintained by Vitko, Inc., which sells one of the products measured.

Burst workload, ordered by median time

Burst of 20 jobs
ConfigurationMedian timep95 timeQueue p50Queue p95Cost for all 20 jobsOK
Blacksmith97 s190 s9.0 s16.0 s$0.16630 of 30
Vitko Runners (warm starts + test splitting)Vitko131 s191 s17.3 s115.1 s$0.05230 of 30
Self-hosted runners on Google Cloud153 s165 s2.0 s2.5 s$0.13530 of 30
GitHub-hosted larger runner (4 vCPU)154 s179 s3.0 s38.6 s$0.63630 of 30
Vitko RunnersVitko160 s372 s18.0 s294.6 s$0.05630 of 30
Self-hosted runners on AWS (one VM per job)161 s182 s65.3 s92.6 s$0.11130 of 30
RunsOn177 s340 s22.5 s193.6 s$0.05130 of 30
Ubicloud203 s271 s18.0 s127.5 s$0.06230 of 30
GitHub-hosted runners (ubuntu-latest, 2 vCPU)259 s283 s3.0 s31.3 s$0.50030 of 30

Time and cost cover the jobs of the burst; queue time is per job. Vitko Runners and the other managed services ran at default capacity with nothing sized for the benchmark. Workload details: burst workload page. List price per minute is not cost per job; faster runners finish sooner.

More: all questions answered · 12 options compared · ranked by speed · ranked by cost per job · burst workload.

Method: how the runs were made; data: results.json; every raw record: the repository. Maintained by Vitko, Inc., which sells one of the products measured. Page dated October 28, 2026.