# Avrea is the fastest GitHub Actions runner in RunsOn's benchmark

Avrea leads RunsOn's 4-vCPU x64 build benchmark, completing Rust, TypeScript and Docker workloads 163% faster than GitHub. See all provider results.

[All posts](https://avrea.com/blog)

[Blog](https://avrea.com/blog)
[CI/CD](https://avrea.com/category/ci-cd)

Avrea leads RunsOn's 4-vCPU x64 build benchmark, completing Rust, TypeScript and Docker workloads 163% faster than GitHub. See all provider results.

[Hannu Valtonen](https://avrea.com/team/hannu-valtonen)
3 November 2026

In [RunsOn's four-vCPU x64 benchmark](https://runs-on.com/benchmarks/runners/), Avrea completed the Rust, TypeScript and Docker workloads in 6m58s. GitHub took 18m18s, so Avrea was 163% faster across the three builds.

A lot of work went into those six minutes and fifty-eight seconds, and we’re proud of our team that made it happen.

Avrea runs your GitHub Actions jobs in virtual machines on dedicated hardware. Package and build caches sit alongside the runners, and your workflows stay in GitHub Actions.

## The full benchmark results

RunsOn tests three public projects with build caching turned off: a [PostHog](https://github.com/PostHog/posthog) Docker build, [opencode](https://github.com/anomalyco/opencode)’s TypeScript checks and [spotify-player](https://github.com/aome510/spotify-player)’s Rust tests. The ranking adds the median build times for all three.

The table shows each provider’s fastest complete configuration in the four-vCPU x64 category.

4 vCPU · x64 · 9 providers · [View all configurations on RunsOn ↗](https://runs-on.com/benchmarks/runners/)

| RunnerWorkspace · CPU | Cost for all buildsUSD ↓ | Time for all builds↓ Lower is faster | Builds, wall time | CPU | Workspace disk | Net & cache | DB | Queue & price | | | | | |
| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |
| Rust↓ | TypeScript↓ | Docker↓ | 1 threadPassMark ↑ | Rand writeIOPS ↑ | QD1 readIOPS ↑ | GitHub DLMiB/s ↑ | Cache restore4 GB ↓ | pgbenchTPS ↑ | QueueBurst ↓ | PriceUSD/min ↓ | | | |
| [Avrea](https://runs-on.com/benchmarks/runners/avrea/) | Dedicated hardware · OVH (FR) | | | | | | | | | | | | |
| [4 vCPU](https://runs-on.com/benchmarks/runners/avrea/#avrea-x64)4 vCPU · 16 GB · VM disk EPYC 4585PX | $0.0625 | 6m 58s | 1m 26s$0.0127 | 40s$0.0070 | 4m 52s$0.0428 | 4,568 | 172k | 50k | 167 | 11s | 9.3k | 11sp90 10m 10s | $0.0080 |
| Namespace | Namespace Labs (US) | | | | | | | | | | | | |
| 4x164 vCPU · 16 GB · VM disk EPYC · model hidden | $0.0904 | 7m 01s | 1m 31s$0.0194 | 50s$0.0122 | 4m 40s$0.0588 | 4,627 | 560k | 105k | 219 | 18s | 13k | 10sp90 7m 00s | $0.0120* |
| RunsOn | EC2 spot rates; on-demand rate shown as OD | | | | | | | | | | | | |
| r8a.xlarge4 vCPU · 32 GB · tmpfs EPYC 9R45 | $0.0182 | 7m 38s | 1m 51s$0.0043 | 50s$0.0025 | 4m 57s$0.0115 | 3,825 | 3.8M | 1.1M | 188 | 5s | 21k | 16sp90 19s | $0.0019OD $0.0056 |
| StarSling | Latitude.sh / AWS · US | | | | | | | | | | | | |
| 4 vCPU4 vCPU · 16 GB · VM disk Xeon · model hidden +3 | $0.0683 | 7m 50s | 1m 32s$0.0131 | 47s$0.0078 | 5m 31s$0.0475 | 3,296 | 68k | 30k | 180 | 21s | 11k | 9sp90 13m 50s | $0.0080 |
| Blacksmith | Bare metal · Phoenix, AZ | | | | | | | | | | | | |
| 4 vCPU4 vCPU · 16 GB · VM disk EPYC · model hidden | $0.0747 | 8m 06s | 1m 40s$0.0148 | 1m 25s$0.0163 | 5m 01s$0.0436 | 4,297 | 218k | 71k | 83 | 12s | 13k | 10sp90 13s | $0.0080 |
| Warpbuild | WarpBuild / AWS / MacStadium · US | | | | | | | | | | | | |
| 4 vCPU4 vCPU · 16 GB · VM disk Ryzen 9 9950X +2 | $0.0908 | 10m 43s | 2m 09s$0.0180 | 1m 07s$0.0108 | 7m 27s$0.0620 | 4,286 | 71k | 30k | 113 | 24s | 3.2k | 21sp90 13m 29s | $0.0080 |
| Ubicloud | Hetzner / Leaseweb · Germany | | | | | | | | | | | | |
| standard 4 vCPU4 vCPU · 16 GB · VM disk EPYC 9454P +3 | $0.0398 | 14m 31s | 2m 19s$0.0064 | 1m 35s$0.0047 | 10m 37s$0.0288 | 2,897 | 27k | 9.0k | 46 | 1m 10s | 2.0k | 40sp90 19m 44s | $0.0025 |
| [GitHub](https://runs-on.com/benchmarks/runners/github/) | Azure · US, CA, MX | | | | | | | | | | | | |
| [4-core](https://runs-on.com/benchmarks/runners/github/#github-x64)4 vCPU · 16 GB · network disk EPYC 7763 +5 | $0.2612 | 18m 18s | 3m 33s$0.0446 | 1m 52s$0.0290 | 12m 53s$0.1876 | 3,398 | 10k | 7.2k | 100 | 44s | 5.0k | 3sp90 5s | $0.0120 |
| AWS CodeBuild | The TypeScript test failed; the Docker workload was not run. | | | | | | | | | | | | |
| medium4 vCPU · 8 GB · EBS Xeon 8488C +1 | Not available | Not available | 3m 32s$0.0380 | [Out of memory](https://github.com/runs-on-demo/benchmark-public/actions/runs/41770952429/job/126380314295) | Not run | 3,004 | 3.0k | 1.8k | 233 | 36s | Not measured | 21sp90 16m 12s | $0.0100 |

Source: [RunsOn’s runner benchmark](https://runs-on.com/benchmarks/runners/), with x64, 4 vCPU and Time selected. AWS CodeBuild has no complete result.

Build costs appear below each workload’s time. RunsOn calculates them from full job duration and the applicable per-minute rate. Its own rows use EC2 spot rates; OD is the on-demand rate. *Namespace uses its overage rate. Queue results show the median, with p90 below.

RunsOn publishes the [methodology](https://runs-on.com/benchmarks/methodology/), [benchmark code and job logs](https://github.com/runs-on-demo/benchmark-public), so you can inspect how each workload was run.

RunsOn estimates a total cost of $0.0625 on Avrea and $0.2612 on GitHub for the three builds. Avrea is 76% cheaper in this test. Avrea charges $0.008 a minute for this runner and GitHub $0.012.

[Try Avrea free](https://console.avrea.com/?utm_source=avrea-blog&utm_medium=blog&utm_campaign=runson-benchmark&utm_content=table-button)

## Eight minutes off a production Docker build

The PostHog Docker workload took 4m52s on Avrea and 12m53s on GitHub.

The job builds PostHog’s frontend and plugin server, installs Python dependencies and assembles the production image. RunsOn uses a fixed revision of PostHog’s Dockerfile and disables the Docker build cache, so every layer is built on each run.

Most of the difference came from building the image: 4m25s on Avrea and 12m22s on GitHub. The full workload also includes fetching the source and setting up Buildx. The [Avrea](https://runs-on.com/benchmarks/runners/avrea/#avrea-x64) and [GitHub](https://runs-on.com/benchmarks/runners/github/#github-x64) profiles publish the timings for each phase.

## Faster TypeScript and Rust builds

The TypeScript workload runs against opencode. It clones the repository, installs dependencies with Bun and runs the monorepo’s typechecks.

Avrea ran the workload 2.8 times as fast as GitHub, 40 seconds against 1m52s.

Turborepo’s local and remote task caches are switched off, so each package is typechecked on every run. Concurrency is set to four for all providers.

For Rust, RunsOn uses spotify-player, a terminal Spotify client. The job installs system dependencies and the Rust toolchain, checks formatting, runs tests and runs Clippy. The benchmark uses no build cache, so every crate is compiled from scratch.

Avrea finished the job in 1m26s and GitHub in 3m33s.

## Your own builds can use the caches

For a Turborepo project, the runner environment is [already configured for the remote cache](https://docs.avrea.com/cache/build-cache/turborepo/), so tasks whose inputs haven’t changed are restored instead of run again.

Rust, C and C++ builds can use [sccache](https://docs.avrea.com/cache/build-cache/sccache/). The backend is already configured on Avrea runners: install sccache and set `RUSTC_WRAPPER: sccache`, and crates that haven’t changed come from the cache instead of being compiled again.

Dependency downloads use a separate cache. Avrea’s [colocated package cache](https://docs.avrea.com/cache/packages/npm/) supports npm, pnpm, Yarn Classic and Bun. Those package managers are preconfigured on Avrea runners to fetch previously cached packages from storage beside the machines running your jobs.

## Run your next build on Avrea

[Sign up for Avrea](https://console.avrea.com/?utm_source=avrea-blog&utm_medium=blog&utm_campaign=runson-benchmark&utm_content=closing-link), install the GitHub App for your repository, and make a copy of a build or test workflow. Then change its runner label to the [Avrea runner size](https://docs.avrea.com/runners/) that fits the workflow. The label for the four-vCPU x64 runner tested here is:

```
runs-on: avrea-ubuntu-latest-4-vcpu
```

Compare both workflows on the same commit, then run the Avrea workflow again to see whether caching shortens the build. Eligible organizations receive a monthly credit equivalent to 3,000 minutes on 2-vCPU Linux x64 runners. Other metered usage draws on the same credit; see the [billing guide](https://docs.avrea.com/billing/).

[Try Avrea free](https://console.avrea.com/?utm_source=avrea-blog&utm_medium=blog&utm_campaign=runson-benchmark&utm_content=closing-button)

## Explore other articles

- 10 Jul 2026
The Fastest GitHub Actions Runners in 2026: CPU Benchmarks Are Half the Story
Read
<https://avrea.com/blog/fastest-github-actions-runners>

- 23 Oct 2026
Avrea is the fastest ARM runner for GitHub Actions: 2026 benchmark
Read
<https://avrea.com/blog/fastest-arm-github-actions-runners>

- 30 Jun 2026
GitHub Actions runners: hosted, self-hosted and managed
Read
<https://avrea.com/blog/github-actions-runners>

## Start shipping with Avrea

[Start free](https://console.avrea.com/)
[Talk to an engineer](https://avrea.com/demo)
