WarpBuild alternatives for GitHub Actions runners

Which alternative to pick

• Avrea, if build time matters most. It finished all four builds in RunsOn’s November benchmark sooner than WarpBuild, and charges WarpBuild’s 2-vCPU x86 rate.
• RunsOn, if runners have to stay in your own AWS account, as WarpBuild’s BYOC runners do.
• Ubicloud, if the per-minute rate matters most.
• Depot or Namespace, if you want to keep a remote Docker builder.
• GitHub’s larger runners, if you’d rather keep CI with one vendor.

Prices for a 2-vCPU Linux runner

Prices checked 11 November 2026.

Avrea, Blacksmith and WarpBuild charge $0.004 a minute for a 2-vCPU x86 runner with 8 GB, and Namespace charges the same within a paid plan. ARM prices vary more, partly because the hardware does.

Prices are per minute for a 2-vCPU Linux runner with 8 GB; Blacksmith’s and Ubicloud’s ARM runners have 6 GB. WarpBuild’s BYOC runners cost $0.002 a minute plus your own cloud bill. Depot’s plans are $20 for one user and $200 for unlimited users. Namespace charges $0.006 beyond a paid plan’s included minutes, and its $100 Team plan’s 100,000 unit minutes cover 25,000 minutes on 2-vCPU, 8-GB runners. Ubicloud’s premium x86 runners cost $0.002.

Avrea’s ARM runners are Apple M5 Max machines. WarpBuild also sells Linux ARM runners on Apple silicon, from 6 vCPU at $0.08 a minute, and Namespace has them in early access; WarpBuild’s standard ARM runners and the others here use server processors such as Graviton4, AmpereOne and Ampere Altra. The build results below show what that difference does to job time, and a job that finishes sooner bills fewer minutes.

Rates exclude storage, add-ons and enterprise terms. Each provider converts included usage into runner minutes at its own ratio. Avrea’s monthly credit is equivalent to 3,000 minutes on 2-vCPU Linux x86 runners and is shared with other metered usage, including persistent disks and static IPs. Avrea and Depot meter jobs by the second; GitHub rounds each job up to a whole minute. Standard GitHub-hosted runners are free in public repositories.

On macOS, Avrea’s smallest runner has 8 vCPU and 24 GB on Apple M5 Max for $0.08 a minute. WarpBuild (M4 Pro or M5 Pro) and Blacksmith (M4) charge the same for 6 vCPU, and Depot for 8 vCPU. Namespace starts at $0.04 for 4 vCPU within a plan, and GitHub’s standard macOS runner is $0.062. Ubicloud and RunsOn have no macOS runners.

Sources: WarpBuild pricing, Avrea pricing, Blacksmith pricing and runner specifications, Depot pricing and runner types, Namespace pricing, Ubicloud pricing, RunsOn pricing, GitHub runner rates and included usage.

Build and cache results at 4 vCPU

RunsOn, a runner vendor that also appears in the table, runs a public benchmark that builds the same projects on every provider’s runners, with its workflows and method published. The 4 November 2026 run built two projects from a cold start on 4-vCPU, 16-GB runners: the Rust project spotify-player, with its tests and clippy, and PostHog’s production Docker image. The table shows median build times from five runs, and the median time to restore a 4 GB file through actions/cache in the 5 November cache test. Depot is not in the benchmark.

Avrea finished every build sooner than WarpBuild: 1.5 times faster on both x86 builds, 2.4 times faster on the ARM Rust build and 2.1 times faster on the ARM Docker image. It also had the fastest ARM build of any provider in both projects. On x86, Avrea had the fastest Rust build and Namespace built the Docker image 12 seconds sooner. These builds start without a warm cache, so they measure the machine and its network rather than cache reuse.

WarpBuild’s and Namespace’s cache times use GitHub’s cache service, because their own caches need a different action. Namespace’s ARM row is its AmpereOne runner; its Apple silicon Linux runner, in early access, built the two projects in 81 and 309 seconds. WarpBuild’s ARM row is its standard 4-vCPU runner; its Apple silicon ARM runners start at 6 vCPU. RunsOn’s row is its fastest EC2 instance with an EBS or local NVMe disk in each test, and Ubicloud’s is its standard runner. Where a run failed, the median covers the successful runs.

Data: RunsOn, benchmarks and cache results, CC BY 4.0. We selected the 4-vCPU rows.

The alternatives

Avrea

We build Avrea. It runs GitHub Actions jobs on dedicated machines: Linux x86, Linux ARM and macOS on Apple M5 Max, and Windows Server 2025. At WarpBuild’s x86 rates it finished every build above sooner, and its ARM runner had the fastest ARM builds in the benchmark.

Supported actions/cache calls go to Avrea’s cache without workflow changes, with 25 GB per repository shared across Actions, LFS, build and package caches. Integrations for Turborepo, Gradle, Bazel, Nx, Go and sccache let a cache hit skip compilation or tests, and persistent disks keep Docker builder state between jobs for $0.50 per provisioned GB a month. Each job has searchable logs, CPU and memory metrics and SSH, and workflow analytics track failure and flake rates. Avrea AI, in beta, proposes a fix for a failing job and runs your tests on it before opening a pull request.

Avrea is a hosted service. For teams that need network controls, the Enterprise plan adds an egress firewall and static IPs. Avrea is ISO 27001:2022 certified and SOC 2 Type II attested. Read the full Avrea and WarpBuild comparison.

Blacksmith

Blacksmith runs managed runners with no plan fee and 3,000 free minutes a month, at $0.004 a minute for 2-vCPU x86 and $0.0025 for 2-vCPU ARM. Its Actions cache needs no workflow changes and includes 25 GB per repository at no extra charge. Docker builds run on the runner, with layers kept on a sticky disk through Blacksmith’s own setup and build actions. macOS runs on M4 from 6 vCPU, and jobs run in four regions across the US and Europe. Avrea vs Blacksmith.

Depot

Depot pairs its runners with remote Docker builders. depot build and depot/build-push-action send the build to a remote 16-CPU BuildKit machine with a persistent layer cache, billed at $0.04 a minute after the plan’s 500 or 5,000 included minutes. Its runners charge the same rate for x86 (AMD EPYC) and ARM (Graviton4) and come with a $20 single-user or $200 unlimited-user plan. Depot Cache also serves Bazel, Gradle, Turborepo, Nx and sccache, and its Business plan can deploy runners into your AWS account. Avrea vs Depot.

Namespace

Namespace sells runners, remote Docker builders, cache volumes and devboxes from one account. Prices are set in unit minutes: a 2-vCPU, 8-GB Linux runner costs $0.004 a minute within a paid plan and $0.006 beyond it, and the Team plan is $100 a month. GitHub’s cache keeps working unchanged; Namespace’s NVMe cache volumes need its cache action and a few workflow edits. Linux ARM runs on AmpereOne, with Apple silicon Linux in early access, and Windows needs the Team plan or above. Avrea vs Namespace.

Ubicloud

Ubicloud’s managed runners are part of an open-source cloud platform and have the lowest hosted rate here, $0.00125 a minute for a standard 2-vCPU runner on x86 or ARM. Premium x86 runners cost $0.002, and new accounts get them by default. The transparent cache works without workflow changes and includes 30 GB per repository. Jobs run in European data centres by default, and there are no macOS or Windows runners. Avrea vs Ubicloud.

RunsOn

RunsOn starts a fresh EC2 instance in your own AWS account for each job, from one CloudFormation stack. You pay AWS for the instances and a small control plane, and RunsOn a flat licence from €300 a year for under 50,000 jobs a month; it is free for non-profit, open-source and personal non-commercial projects. RunsOn’s own calculator puts a 4-vCPU spot instance at $0.0014 a minute in us-east-1, before the licence and control plane. Linux and Windows run on x86, ARM and GPU instances, and macOS jobs stay on GitHub’s runners. You manage the AWS account and its capacity, as with WarpBuild’s BYOC runners, but only on AWS.

GitHub-hosted runners

Staying on GitHub keeps CI with one vendor and one bill. Standard 2-core Linux runners are free in public repositories and $0.006 a minute in private ones. Larger runners need a Team or Enterprise Cloud plan, are billed in public repositories too, and cost $0.012 a minute for 4-core x86 and $0.008 for 4-core ARM. Each plan includes $12 to $300 of usage a month. Avrea vs GitHub-hosted runners.

Self-hosted runners with Actions Runner Controller

Actions Runner Controller runs GitHub’s self-hosted runner on your own Kubernetes cluster, in any cloud. GitHub doesn’t charge for self-hosted runner minutes, so the cost is the cluster and the time it takes to run it: images, autoscaling, cache storage and security updates are yours.

When staying on WarpBuild makes sense

WarpBuild is the one provider here that deploys managed runners into AWS, GCP or Azure, at $0.002 a minute plus the cloud bill, with cache and snapshots included. A team that has to keep CI inside its own cloud network, on GCP or Azure in particular, has no hosted equivalent among the others. Its 2-vCPU ARM rate, $0.003 a minute, is below Avrea’s and GitHub’s, it offers Linux ARM on Apple silicon as well as on server processors, and macOS runs on M4 Pro or M5 Pro at 6 and 12 vCPU.

Moving means changing each job’s runner label and, where a workflow uses WarpBuilds/cache, switching it back to actions/cache. The first runs elsewhere start with a cold cache.

Common questions

Which is the fastest WarpBuild alternative?

Avrea, in RunsOn’s 4 November 2026 run on 4-vCPU runners. It had the fastest ARM builds of both projects and the fastest x86 Rust build, and restored the cache fastest of the hosted providers. Namespace built the x86 Docker image 12 seconds sooner than Avrea. Depot is not in the benchmark.

Is Avrea faster than WarpBuild?

In the same run, Avrea finished all four builds sooner than WarpBuild: the x86 Rust build in 86 seconds against 129, the x86 Docker image in 292 against 447, and the ARM builds 2.4 and 2.1 times faster. Both charge $0.004 a minute for a 2-vCPU x86 runner, so on x86 the same job bills fewer minutes on Avrea.

What is the cheapest alternative to WarpBuild?

Per minute, Ubicloud’s standard runners: $0.00125 for 2 vCPU on x86 or ARM. If you use WarpBuild’s BYOC runners on AWS, compare RunsOn, which also bills through your own AWS account, plus its annual licence. Per job, speed counts as well: Avrea charges WarpBuild’s x86 rate and finished the same x86 builds in about two thirds of the time.

Which WarpBuild alternatives run in my own cloud account?

RunsOn runs in your AWS account. Depot offers deployment into your AWS account on its Business plan. Actions Runner Controller runs on any Kubernetes cluster you operate, in any cloud. None of them covers GCP and Azure as a managed service, as WarpBuild’s BYOC runners do. Avrea, Blacksmith, Namespace and Ubicloud are hosted services; Avrea’s Enterprise plan adds an egress firewall and static IPs for teams that need network controls.

Which alternatives have macOS runners?

Avrea runs macOS 26 on Apple M5 Max at 8 and 16 vCPU. Blacksmith runs M4 at 6 vCPU, Depot runs macOS 26 on M4 at 8 vCPU, Namespace offers Apple silicon from 4 vCPU, and GitHub has standard and larger macOS runners. Ubicloud and RunsOn don’t run macOS.

How do I move a workflow off WarpBuild?

Change the job’s runs-on label to the new provider’s, for example avrea-ubuntu-latest for Avrea’s 2-vCPU, 8-GB x86 runner. On Avrea, the console’s migration wizard converts a workflow and opens the pull request for you. If a workflow uses WarpBuilds/cache, switch it back to actions/cache, which Avrea serves from its own cache without further changes. Expect the first runs to rebuild caches from scratch, and compare a cold run, a repeat of the same commit and a typical change before deciding.

Related reading

• Avrea vs WarpBuild: pricing, ARM and macOS hardware, caching and running in your own cloud.
• GitHub Actions runner prices compared: every provider’s rates and the cost of one build.
• Fastest ARM runners for GitHub Actions: the ARM build results in detail.
• All runner comparisons: the wider provider comparison.

Try one job on Avrea

Connect a repository with the setup guide, then change one job’s runner label. Run the same commits on both providers and compare duration and cost, including a cold cache.

Try Avrea · Runner sizes and pricing