Short answer
CI Bench does not publish CPU model or microbenchmark scores; it measures whole CI jobs. On the four compile- and test-heavy jobs below, the lowest median time was: fd: Vitko Runners (warm starts + test splitting) (45 s, GitHub-hosted 92 s); cli: Vitko Runners (warm starts + test splitting) (140 s, GitHub-hosted 368 s); guava: Vitko Runners (warm starts + test splitting) (651 s, GitHub-hosted 1167 s); monorepo: Vitko Runners (warm starts + test splitting) (172 s, GitHub-hosted 384 s). Job time also includes setup, caches, network and disk, so it is not a pure CPU score. For CPU microbenchmarks by provider, RunsOn publishes its own CPU benchmark (linked below). Maintained by Vitko, Inc., which sells one of the products measured.
Median job time on four compile- and test-heavy workloads
| Workload | GitHub-hosted runners (ubuntu-latest, 2 vCPU) | GitHub-hosted larger runner (4 vCPU) | Vitko Runners | Vitko Runners (warm starts + test splitting) | Blacksmith | Ubicloud | RunsOn | Self-hosted runners on AWS (one VM per job) | Self-hosted runners on Google Cloud |
|---|---|---|---|---|---|---|---|---|---|
| fd | 92 s | 66 s | 49 s | 45 s | 65 s | 87 s | 89 s | 114 s | 64 s |
| cli | 368 s | 192 s | 272 s | 140 s | 286 s | 359 s | 327 s | 357 s | 310 s |
| guava | 1167 s | 946 s | 655 s | 651 s | 735 s | 881 s | 897 s | 717 s | 820 s |
| monorepo | 384 s | 220 s | 180 s | 172 s | 177 s | 276 s | 288 s | 281 s | 278 s |
All configurations are 2 vCPU except GitHub-hosted larger runner (4 vCPU). Workloads: fd (Rust build and test), GitHub CLI (Go test and build), guava (Java build and tests), monorepo (vercel/ai build). Every workload page lists p95, queue time and cost: all workloads.
What providers publish about CPUs
| Provider | What its page says | Source |
|---|---|---|
| Depot | Intel x86 runners use 4th Gen AMD EPYC Genoa CPUs; Arm runners use AWS Graviton4. | Depot docs, overview, read 2026-10-27 |
| Ubicloud | Runners come with dedicated CPU, memory and local block storage on bare-metal providers such as Hetzner, Leaseweb or AWS bare metal. | Ubicloud pricing, read 2026-10-27 |
| RunsOn | Runs on EC2 instances in your AWS account; you choose the instance family and size by label, for example 8th-generation general-purpose instances. | RunsOn, read 2026-10-27 |
For CPU microbenchmarks (rather than whole jobs), see RunsOn’s published CPU benchmark, a third-party source we have not verified; RunsOn is one of the providers CI Bench measures.
More: 12 options compared · ranked by speed · ranked by cost per job · price per minute, 2 and 4 vCPU · self-hosted vs managed · alternatives to GitHub-hosted.
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 27, 2026.