Avrea is the fastest ARM runner for GitHub Actions: 2026 benchmark
Avrea has the fastest ARM runners for GitHub Actions, according to RunsOn's independent benchmark. In the October runs, 36 arm64 runners from seven providers built the same Rust project from a cold start, and Avrea's Apple M5 Max runners finished first at every size tested: 118 seconds at 2 vCPU, 80 at 4 and 61 at 8. GitHub's own arm64 runners took 314 and 190 seconds at 2 and 4 vCPU.
Where a provider has several runners of one size, the table shows its fastest. Namespace's Apple silicon runners are an early-access option on M4 Pro or M5 Max hosts. WarpBuild's Apple M4 Pro runners, at 6 and 12 vCPU, took 86 and 65 seconds.
Against GitHub's runner of the same size, the job ran 2.7x faster on Avrea at 2 vCPU and 2.4x at 4 vCPU. At 2 vCPU it also cost about the same at list prices, $0.026 on Avrea against $0.027 on GitHub: the per-minute price is 2.4x GitHub's and the job finished 2.7x faster.
What RunsOn measured
RunsOn sells self-hosted GitHub Actions runners for your own AWS account, and it also runs a public benchmark of everyone's runners: 93 runner labels from nine providers build the same three open-source projects from a cold start every few days, with every job left on GitHub for anyone to read. The Rust job is a cold build of spotify-player, a Rust terminal client: checkout, apt packages, toolchain install, cargo fmt, cargo test and two cargo clippy passes, with no cache. Times are medians of five runs between 10 and 22 October (four for one Namespace runner), in seconds of build time.
Why the M5 Max is this far ahead
A cargo build has a long serial path: crates wait on their dependencies, and much of each compile runs on one core, so the speed of one core lands directly on the wall clock. RunsOn's hardware suite puts the M5 Max at 5,028 PassMark single-thread points on the 2 vCPU runner. Next is Namespace's Apple silicon runner at 4,750, then Graviton5 at about 2,470, the Neoverse-V2 runners at about 1,940, GitHub's Neoverse-N2 at 1,874, and the Ampere and Neoverse-N1 runners between 1,170 and 1,340.
The phase timings say the same thing. cargo test took 49 seconds on Avrea's 4 vCPU runner and 114 on GitHub's. The main clippy pass took 16 seconds against 46. More cores help less than a faster core: Avrea's 8 vCPU runner finished in 61 seconds, Blacksmith's in 184 and Namespace's Ampere runner in 233. The same M5 Max chips run Avrea's macOS runners; the launch post has the sizes for both.
Run the same job
RunsOn's workflows are public in runs-on-demo/benchmark-public and the data is CC BY 4.0, so you can check every number above against the job logs. To run your own ARM job on Avrea, install the GitHub App and change the label:
ARM sizes run from 1 to 16 vCPU on Ubuntu 24.04 and 26.04, and the free tier covers about 500 minutes at this size, so a first test costs nothing. The runner labels and billing pages have the full tables. For the x64 half of the same benchmark, and why cache speed decides warm builds, see the fastest GitHub Actions runners.
Frequently asked questions
Which ARM runner is fastest for GitHub Actions?
On RunsOn's independent benchmark, October 2026, Avrea's Apple M5 Max runners finished a cold Rust build first at 2, 4 and 8 vCPU: 118, 80 and 61 seconds, against 314 and 190 seconds on GitHub's arm64 runners. They also posted the highest arm64 single-thread score in the hardware suite, 5,028 PassMark points.
How much faster is Avrea than GitHub's ARM runners?
2.7x at 2 vCPU and 2.4x at 4 vCPU on the same cold Rust build, as measured by RunsOn. At 2 vCPU the job cost about the same on both at list prices, $0.026 on Avrea against $0.027 on GitHub.
How do I run GitHub Actions jobs on Avrea's ARM runners?
Install the Avrea GitHub App and set runs-on to avrea-ubuntu-latest-arm-4-vcpu, or the 1, 2, 8 or 16 vCPU size. ARM images are Ubuntu 24.04 and 26.04, and the free tier covers about 500 minutes at 4 vCPU.
Data: RunsOn GitHub Actions runner benchmarks, runs between 8 and 22 October 2026, published under CC BY 4.0.
