Fastest GitHub Actions Runners in 2026

This guide compares Linux GitHub Actions runner CPU performance, queue times, and pricing across official GitHub runners, self-hosted solutions, and third-party providers. Real-world benchmarks show how each option performs to help select the optimal runner.

Benchmarks cover GitHub, RunsOn, AWS CodeBuild, Namespace, Blacksmith, Cirrus, Warpbuild, Ubicloud, and Shipfox across Linux x64 and arm64.

Note

BuildJet was removed from this leaderboard after the service shut down in January 2026. See BuildJet Is Shutting Down. What Should GitHub Actions Teams Do Next?.

TL;DR

Best value

RunsOn, Namespace, Shipfox, Cirrus, and Blacksmith score highly in this benchmark. Each one having specific strengths and weaknesses depending on your use cases.

Best self-hosted

RunsOn is the cheapest self-hosted option and the fastest in this dataset among self-hosted runners. AWS CodeBuild is slower and more expensive.

Fastest CPU

x64: Shipfox, Namespace, Blacksmith, and RunsOn lead the pack. arm64: Cirrus and Namespace lead, followed by RunsOn, Shipfox, and Warpbuild.

Lowest queue

GitHub is lowest at ~8s (x64) and ~6s (arm64). Warpbuild and Namespace are close behind on x64.

How to read: leaderboard and deep dive use the configured CPU/queue percentiles (p50/p50). Higher CPU is better, lower queue is better. Price is normalized $/min.

GitHub Actions runner leaderboard

Last updated: Feb 19, 2026

Benchmarks use Linux runners only, the Passmark single-thread metric, and the last 30 days of data.

Linux x64 / amd64

Best valueFastest CPULowest queueCheapestSelf-hosted only

Settings

Preference: Cheaper (30) ↔ Faster (70)CheaperFaster

Force on-demand prices for RunsOn

How scoring works

baseScore = (w_speed × speedScore + w_cost × costScore) / (w_speed + w_cost)
speedScore = (cpu - cpuLow) / (cpuHigh - cpuLow), costScore = (priceHigh - price) / (priceHigh - priceLow) (both clamped 0-1)
queueAdj% = -min(15, max(0, queue - queueMin) × 0.5)
finalScore = baseScore × (1 + queueAdj% / 100)

RunsOn

81

✨ runs-on,family=m8azn

CPU p50: 4267

Queue p50: 25.0s

Price: $0.0015/min

Self-Hosted

Namespace

79

nscloud-ubuntu-24.04-amd64-2x8

CPU p50: 4515

Queue p50: 11.0s

Price: $0.0040/min

SaaS

Cirrus

79

ghcr.io/cirruslabs/ubuntu-runner-amd64:24.04-xs, ghcr.io/cirruslabs/ubuntu-runner-amd64:24.04-md

CPU p50: 3958

Queue p50: 17.0s

Price: $0.0010/min

SaaS

CPU speed vs price per minutex64 runners

RunsOnGitHubAWS CodeBuildBlacksmithCirrusNamespaceShipfoxUbicloudWarpbuild

Bigger dots = longer queues. Hover for details.

Provider	CPU p50	Queue p50	Price / min	Value score
RunsOn ✨ runs-on,family=m8azn AWS Self-Hosted spot	4267	25.0s	$0.0015/min	81
Namespace nscloud-ubuntu-24.04-amd64-2x8 US/Europe SaaS standard	4515	11.0s	$0.0040/min	79
Cirrus ghcr.io/cirruslabs/ubuntu-runner-amd64:24.04-xs, ghcr.io/cirruslabs/ubuntu-runner-amd64:24.04-md Hetzner SaaS per runner monthly	3958	17.0s	$0.0010/min	79
Shipfox shipfox-2vcpu-ubuntu-2404 OVH / GCP SaaS	4539	16.0s	$0.0040/min	78
Blacksmith blacksmith-2vcpu-ubuntu-2404 Hetzner SaaS	4484	18.0s	$0.0040/min	76
RunsOn ✨ runs-on,family=m8a AWS Self-Hosted spot	3820	26.0s	$0.0009/min	72
Ubicloud ubicloud-premium-2-ubuntu-2404 Hetzner (Germany) SaaS premium	3709	30.0s	$0.0016/min	65
Warpbuild warp-ubuntu-2404-x64-2x Hetzner / AWS / BYOC SaaS	3731	12.0s	$0.0040/min	59
RunsOn ✨ runs-on,family=m8i+r8i AWS Self-Hosted spot	3195	28.0s	$0.0009/min	57
Ubicloud ubicloud-standard-2-ubuntu-2404 Hetzner (Germany) SaaS standard	2903	17.0s	$0.0008/min	54
RunsOn ✨ runs-on,family=r7iz AWS Self-Hosted spot	3070	28.0s	$0.0011/min	53
RunsOn ✨ runs-on,family=c7i+m7i+r7i AWS Self-Hosted spot	2841	28.0s	$0.0009/min	48
RunsOn ✨ runs-on,family=c7a+m7a+r7a AWS Self-Hosted spot	2878	28.0s	$0.0011/min	48
RunsOn ✨ runs-on,family=i7ie AWS Self-Hosted spot	2729	31.0s	$0.0016/min	42
RunsOn ✨ runs-on,family=c6a+m6a+r6a AWS Self-Hosted spot	2588	32.0s	$0.0010/min	41
GitHub ubuntu-24.04, github24-2cpu-x64 Azure Official	2327	7.0s	$0.0060/min	11
GitHub github24-2cpu-x64, ubuntu-24.04 Azure Official	2281	7.0s	$0.0060/min	10
GitHub ubuntu-24.04, github24-2cpu-x64 Azure Official	2203	9.0s	$0.0060/min	8
AWS CodeBuild codebuild-ubuntu-8.0-large, codebuild-ubuntu-8.0-small, codebuild-ubuntu-8.0-medium AWS Self-Hosted	1986	27.0s	$0.0050/min	7
AWS CodeBuild codebuild-ubuntu-8.0-medium, codebuild-ubuntu-8.0-large, codebuild-ubuntu-8.0-small AWS Self-Hosted	1913	28.0s	$0.0050/min	5

Note

Pricing is pulled from each provider’s pricing page and normalized to $/min.

• RunsOn prices here use real EC2 .large family prices in us-east-1 (spot), with EBS + License (EUR 300/yr @ 25d * 12h / 100 jobs/day) added.
• Ubicloud pricing follows the benchmark runner tier (standard or premium)
• Namespace arm64 pricing follows runner variant ($0.004/min standard, $0.012/min with linux-on-apple-silicon=true)
• Cirrus is modeled as a flat monthly price at 25 days × 12 hours, normalized to $/min and the resource class used (xs or md). You still pay the full runner price even if you don’t use all the capacity. Concurrency is limited by how many full runners you buy.

Linux arm64 / aarch64

Best valueFastest CPULowest queueCheapestSelf-hosted only

Settings

Preference: Cheaper (30) ↔ Faster (70)CheaperFaster

Force on-demand prices for RunsOn

How scoring works

baseScore = (w_speed × speedScore + w_cost × costScore) / (w_speed + w_cost)
speedScore = (cpu - cpuLow) / (cpuHigh - cpuLow), costScore = (priceHigh - price) / (priceHigh - priceLow) (both clamped 0-1)
queueAdj% = -min(15, max(0, queue - queueMin) × 0.5)
finalScore = baseScore × (1 + queueAdj% / 100)

Cirrus

91

ghcr.io/cirruslabs/ubuntu-runner-arm64:24.04-md

CPU p50: 3968

Queue p50: 17.0s

Price: $0.0021/min

SaaS

Namespace

67

nscloud-ubuntu-24.04-arm64-2x8,namespace-features:linux-on-apple-silicon=true

CPU p50: 3984

Queue p50: 15.0s

Price: $0.0120/min

SaaS

RunsOn

42

✨ runs-on,family=c8g+m8g+r8g

CPU p50: 1946

Queue p50: 29.0s

Price: $0.0008/min

Self-Hosted

CPU speed vs price per minutearm64 runners

RunsOnGitHubBlacksmithCirrusNamespaceShipfoxUbicloudWarpbuild

Bigger dots = longer queues. Hover for details.

Provider	CPU p50	Queue p50	Price / min	Value score
Cirrus ghcr.io/cirruslabs/ubuntu-runner-arm64:24.04-md Hetzner SaaS per runner monthly	3968	17.0s	$0.0021/min	91
Namespace nscloud-ubuntu-24.04-arm64-2x8,namespace-features:linux-on-apple-silicon=true US/Europe SaaS premium	3984	15.0s	$0.0120/min	67
RunsOn ✨ runs-on,family=c8g+m8g+r8g AWS Self-Hosted spot	1946	29.0s	$0.0008/min	42
Shipfox shipfox-2vcpu-ubuntu-2404-arm OVH / GCP SaaS	2084	20.0s	$0.0040/min	40
Warpbuild warp-ubuntu-latest-arm64-2x, warp-ubuntu-2404-arm64-2x Hetzner / AWS / BYOC SaaS	1944	37.0s	$0.0030/min	36
GitHub github24-2cpu-arm64 Azure Official	1874	6.0s	$0.0050/min	35
RunsOn ✨ runs-on,family=c7g+m7g+r7g AWS Self-Hosted spot	1559	29.0s	$0.0008/min	33
Ubicloud ubicloud-standard-2-arm-ubuntu-2404 Hetzner (Germany) SaaS standard	1315	15.0s	$0.0008/min	30
Blacksmith blacksmith-2vcpu-ubuntu-2404-arm Hetzner SaaS	1328	19.0s	$0.0025/min	26
Namespace nscloud-ubuntu-24.04-arm64-2x8 US/Europe SaaS standard	1251	15.0s	$0.0040/min	20
GitHub github24-2cpu-arm64 Azure Official	1333	12.0s	$0.0050/min	20

Note

Pricing uses provider public rates. Some ARM offers (Apple virtualization, premium tiers) can vary by region and plan.

How to choose a GitHub Actions runner

• If you want the lowest price, compare $/min.
• If you want speed, look at the configured CPU percentile (currently p50 single‑thread).
• If you want the lowest queue, look at the configured queue percentile (currently p50). GitHub is usually best, followed by Warpbuild and Namespace.
• If you want predictability, favor providers with stable hardware (some Hetzner‑based fleets can vary CPUs).
• If you want control, pick self‑hosted (RunsOn, AWS CodeBuild) and keep data in your cloud account. RunsOn keeps runners inside your AWS VPC; see networking options.

Providers included

See full provider list

Official:

• GitHub (Azure)

Self-hosted alternatives:

• 👋 RunsOn (AWS)
• AWS CodeBuild (AWS)

Third-party alternatives:

• Namespace (US/Europe)
• Blacksmith (Hetzner / Secured Servers LLC)
• Cirrus (Hetzner)
• Shipfox (OVH, GCP)
• Warpbuild (Hetzner / AWS / BYOC)
• Ubicloud (Hetzner)

Note: I’d love to benchmark Depot (hosted on AWS), but they forbid benchmarking their platform 🤷.

Deep dive (full data)

Open full benchmark table & box plots

Includes per-runner distributions, CPU/queue percentiles, and provider metadata for the last 30 days.

Linux x64 / amd64

Provider

CPU speed (p50)
(higher is better)

Queue time (p50)
(lower is better)

Processor

Shipfox

SaaS $0.004/min 1.5x cheaper

• shipfox-2vcpu-ubuntu-2404

CPU speed

p50: 4539p90: 4607

Queue time

p50: 16.0sp90: 20.0s

AMD EPYC 4585PX 16-Core Processor(x86_64)

• Infra / ISP: OVH / GCP
• Samples: 31

Namespace

SaaS $0.004/min 1.5x cheaper standard

• nscloud-ubuntu-24.04-amd64-2x8

CPU speed

p50: 4507p90: 4672

Queue time

p50: 11.0sp90: 13.0s

AMD EPYC (x86_64)

• Infra / ISP: US/Europe (might vary)
• Price basis: Standard Linux VM (2 vCPU / 8GB + caching)
• Samples: 32

---

CPU 4500+

---

Blacksmith

SaaS $0.004/min 1.5x cheaper

• blacksmith-2vcpu-ubuntu-2404

CPU speed

p50: 4479p90: 4622

Queue time

p50: 16.0sp90: 24.0s

AMD EPYC (x86_64)

• Infra / ISP: Hetzner
• Samples: 32

RunsOn

Self-Hosted $0.0015/min 4.0x cheaper spot

• runs-on,family=m8azn

CPU speed

p50: 4260p90: 4273

Queue time

p50: 25.0sp90: 29.0s

AMD EPYC 9R05 (x86_64)

• Infra / ISP: AWS
• Price basis: EC2 m8azn.large (us-east-1, family m8azn, spot default) + EBS (40GB gp3, 400 MB/s) + License (EUR 300/yr @ 25d * 12h / 100 jobs/day)
• Samples: 10

---

CPU 4000+

---

Cirrus

SaaS $0.001/min 21.1x cheaper per runner monthly

• ghcr.io/cirruslabs/ubuntu-runner-amd64:24.04-xs
• ghcr.io/cirruslabs/ubuntu-runner-amd64:24.04-md

CPU speed

p50: 3947p90: 4089

Queue time

p50: 17.0sp90: 20.0s

AMD Ryzen 9 7950X3D 16-Core Processor(x86_64)

• Infra / ISP: Hetzner
• Pricing model: per runner monthly
• Price basis: $18.75/month @ 25d * 12h utilization
• Samples: 32

RunsOn

Self-Hosted $0.0009/min 6.3x cheaper spot

• runs-on,family=m8a

CPU speed

p50: 3820p90: 3829

Queue time

p50: 26.0sp90: 28.0s

AMD EPYC 9R45 (x86_64)

• Infra / ISP: AWS
• Price basis: EC2 m8a.large (us-east-1, family m8a, spot default) + EBS (40GB gp3, 400 MB/s) + License (EUR 300/yr @ 25d * 12h / 100 jobs/day)
• Samples: 37

Warpbuild

SaaS $0.004/min 1.5x cheaper

• warp-ubuntu-2404-x64-2x

CPU speed

p50: 3645p90: 3872

Queue time

p50: 11.0sp90: 86187.0s

AMD Ryzen 9 7950X3D 16-Core Processor(x86_64)

• Infra / ISP: Hetzner / AWS / BYOC
• Samples: 8

Ubicloud

SaaS $0.0016/min 3.8x cheaper premium

• ubicloud-premium-2-ubuntu-2404

CPU speed

p50: 3696p90: 3850

Queue time

p50: 29.0sp90: 37.0s

AMD Ryzen 9 7950X3D 16-Core Processor(x86_64)

• Infra / ISP: Hetzner (Germany)
• Samples: 32

---

CPU 3500+

---

RunsOn

Self-Hosted $0.0009/min 6.5x cheaper spot

• runs-on,family=m8i+r8i

CPU speed

p50: 3195p90: 3206

Queue time

p50: 28.0sp90: 39.0s

Intel(R) Xeon(R) 6975P-C (x86_64)

• Infra / ISP: AWS
• Price basis: EC2 m8i.large (us-east-1, family m8i, spot default) + EBS (40GB gp3, 400 MB/s) + License (EUR 300/yr @ 25d * 12h / 100 jobs/day)
• Samples: 35

RunsOn

Self-Hosted $0.0011/min 5.3x cheaper spot

• runs-on,family=r7iz

CPU speed

p50: 3070p90: 3076

Queue time

p50: 28.0sp90: 37.0s

Intel(R) Xeon(R) Gold 6455B (x86_64)

• Infra / ISP: AWS
• Price basis: EC2 r7iz.large (us-east-1, family r7iz, spot default) + EBS (40GB gp3, 400 MB/s) + License (EUR 300/yr @ 25d * 12h / 100 jobs/day)
• Samples: 23

---

CPU 3000+

---

Ubicloud

SaaS $0.0008/min 7.5x cheaper standard

• ubicloud-standard-2-ubuntu-2404

CPU speed

p50: 2901p90: 2957

Queue time

p50: 17.0sp90: 34.0s

AMD EPYC 9454P 48-Core Processor(x86_64)

• Infra / ISP: Hetzner (Germany)
• Samples: 32

RunsOn

Self-Hosted $0.0011/min 5.4x cheaper spot

• runs-on,family=c7a+m7a+r7a

CPU speed

p50: 2878p90: 2881

Queue time

p50: 28.0sp90: 64.0s

AMD EPYC 9R14 (x86_64)

• Infra / ISP: AWS
• Price basis: EC2 m7a.large (us-east-1, family m7a, spot default) + EBS (40GB gp3, 400 MB/s) + License (EUR 300/yr @ 25d * 12h / 100 jobs/day)
• Samples: 29

RunsOn

Self-Hosted $0.0009/min 6.6x cheaper spot

• runs-on,family=c7i+m7i+r7i

CPU speed

p50: 2821p90: 2956

Queue time

p50: 28.0sp90: 64.0s

Intel(R) Xeon(R) Platinum 8488C (x86_64)

• Infra / ISP: AWS
• Price basis: EC2 m7i.large (us-east-1, family m7i, spot default) + EBS (40GB gp3, 400 MB/s) + License (EUR 300/yr @ 25d * 12h / 100 jobs/day)
• Samples: 28

RunsOn

Self-Hosted $0.0016/min 3.8x cheaper spot

• runs-on,family=i7ie

CPU speed

p50: 2729p90: 3005

Queue time

p50: 31.0sp90: 52.0s

Intel(R) Xeon(R) Platinum 8559C (x86_64)

• Infra / ISP: AWS
• Price basis: EC2 i7ie.large (us-east-1, family i7ie, spot default) + EBS (40GB gp3, 400 MB/s) + License (EUR 300/yr @ 25d * 12h / 100 jobs/day)
• Samples: 23

RunsOn

Self-Hosted $0.001/min 5.8x cheaper spot

• runs-on,family=m6a

CPU speed

p50: 2588p90: 2591

Queue time

p50: 32.0sp90: 59.0s

AMD EPYC 7R13 Processor (x86_64)

• Infra / ISP: AWS
• Price basis: EC2 m6a.large (us-east-1, family m6a, spot default) + EBS (40GB gp3, 400 MB/s) + License (EUR 300/yr @ 25d * 12h / 100 jobs/day)
• Samples: 3

---

CPU 2500+

---

GitHub

Official $0.006/min ≈ GitHub

• ubuntu-24.04
• github24-2cpu-x64

CPU speed

p50: 2327p90: 2329

Queue time

p50: 7.0sp90: 21.0s

Intel(R) Xeon(R) Platinum 8370C CPU @ 2.80GHz (x86_64)

• Infra / ISP: Azure
• Samples: 5

GitHub

Official $0.006/min ≈ GitHub

• github24-2cpu-x64
• ubuntu-24.04

CPU speed

p50: 2280p90: 2288

Queue time

p50: 7.0sp90: 9.0s

AMD EPYC 7763 64-Core Processor(x86_64)

• Infra / ISP: Azure
• Samples: 56

GitHub

Official $0.006/min ≈ GitHub

• ubuntu-24.04
• github24-2cpu-x64

CPU speed

p50: 2198p90: 2203

Queue time

p50: 8.0sp90: 9.0s

AMD EPYC 9V74 80-Core Processor(x86_64)

• Infra / ISP: Azure
• Samples: 2

---

CPU 2000+

---

AWS CodeBuild

Self-Hosted $0.005/min 1.2x cheaper

• codebuild-ubuntu-8.0-large
• codebuild-ubuntu-8.0-small
• codebuild-ubuntu-8.0-medium

CPU speed

p50: 1986p90: 1994

Queue time

p50: 27.0sp90: 29.0s

Intel(R) Xeon(R) Platinum 8275CL CPU @ 3.00GHz (x86_64)

• Infra / ISP: AWS
• Samples: 65

AWS CodeBuild

Self-Hosted $0.005/min 1.2x cheaper

• codebuild-ubuntu-8.0-medium
• codebuild-ubuntu-8.0-large
• codebuild-ubuntu-8.0-small

CPU speed

p50: 1912p90: 1918

Queue time

p50: 28.0sp90: 30.0s

Intel(R) Xeon(R) Platinum 8124M CPU @ 3.00GHz (x86_64)

• Infra / ISP: AWS
• Samples: 30

Linux arm64 / aarch64

Provider

CPU speed (p50)
(higher is better)

Queue time (p50)
(lower is better)

Processor

Namespace

SaaS $0.012/min 2.4x pricier premium

• nscloud-ubuntu-24.04-arm64-2x8,namespace-features:linux-on-apple-silicon=true

CPU speed

p50: 3977p90: 4230

Queue time

p50: 14.0sp90: 19.0s

Apple M4 + Virtualization

• Infra / ISP: US/Europe (might vary)
• Price basis: Premium arm64 variant (linux-on-apple-silicon=true)
• Samples: 32

Cirrus

SaaS $0.0021/min 3.8x cheaper per runner monthly

• ghcr.io/cirruslabs/ubuntu-runner-arm64:24.04-md

CPU speed

p50: 3957p90: 4248

Queue time

p50: 16.0sp90: 36.0s

Apple M4 + Virtualization

• Infra / ISP: Hetzner
• Pricing model: per runner monthly
• Price basis: $37.5/month @ 25d * 12h utilization
• Samples: 32

---

CPU 3500+

---

Shipfox

SaaS $0.004/min 1.3x cheaper

• shipfox-2vcpu-ubuntu-2404-arm

CPU speed

p50: 2084p90: 2086

Queue time

p50: 20.0sp90: 52.0s

Neoverse-V2 (aarch64)

• Infra / ISP: OVH / GCP
• Samples: 32

---

CPU 2000+

---

RunsOn

Self-Hosted $0.0008/min 6.0x cheaper spot

• runs-on,family=c8g+m8g+r8g

CPU speed

p50: 1946p90: 1947

Queue time

p50: 29.0sp90: 41.0s

Neoverse-V2 (aarch64)

• Infra / ISP: AWS
• Price basis: EC2 m8g.large (us-east-1, family m8g, spot default) + EBS (40GB gp3, 400 MB/s) + License (EUR 300/yr @ 25d * 12h / 100 jobs/day)
• Samples: 65

Warpbuild

SaaS $0.003/min 1.7x cheaper

• warp-ubuntu-latest-arm64-2x
• warp-ubuntu-2404-arm64-2x

CPU speed

p50: 1944p90: 1946

Queue time

p50: 37.0sp90: 86357.0s

Neoverse-V2 (aarch64)

• Infra / ISP: Hetzner / AWS / BYOC
• Samples: 7

GitHub

Official $0.005/min ≈ GitHub

• github24-2cpu-arm64

CPU speed

p50: 1874p90: 1879

Queue time

p50: 6.0sp90: 9.0s

Neoverse-N2 (aarch64)

• Infra / ISP: Azure
• Samples: 30

RunsOn

Self-Hosted $0.0008/min 5.9x cheaper spot

• runs-on,family=c7g+m7g+r7g

CPU speed

p50: 1559p90: 1564

Queue time

p50: 29.0sp90: 41.0s

Neoverse-V1 (aarch64)

• Infra / ISP: AWS
• Price basis: EC2 m7g.large (us-east-1, family m7g, spot default) + EBS (40GB gp3, 400 MB/s) + License (EUR 300/yr @ 25d * 12h / 100 jobs/day)
• Samples: 33

---

CPU 1500+

---

GitHub

Official $0.005/min ≈ GitHub

• github24-2cpu-arm64

CPU speed

p50: 1333p90: 1333

Queue time

p50: 6.0sp90: 12.0s

Neoverse-N1 (aarch64)

• Infra / ISP: Azure
• Samples: 2

Blacksmith

SaaS $0.0025/min 2.0x cheaper

• blacksmith-2vcpu-ubuntu-2404-arm

CPU speed

p50: 1328p90: 1332

Queue time

p50: 18.0sp90: 33.0s

Neoverse-N1 (aarch64)

• Infra / ISP: Hetzner
• Samples: 32

Ubicloud

SaaS $0.0008/min 6.3x cheaper standard

• ubicloud-standard-2-arm-ubuntu-2404

CPU speed

p50: 1315p90: 1319

Queue time

p50: 15.0sp90: 20.0s

Neoverse-N1 (aarch64)

• Infra / ISP: Hetzner (Germany)
• Samples: 31

Namespace

SaaS $0.004/min 1.3x cheaper standard

• nscloud-ubuntu-24.04-arm64-2x8

CPU speed

p50: 1251p90: 1264

Queue time

p50: 15.0sp90: 18.0s

Ampere-1a (aarch64)

• Infra / ISP: US/Europe (might vary)
• Price basis: Standard arm64 variant (without linux-on-apple-silicon=true)
• Samples: 19

Methodology & data sources

• Leaderboard and deep dive use the configured CPU/queue percentiles (currently p50/p50). CPU is Passmark single-thread; queue is time-to-start in seconds.
• Data window is the last 30 days before the last updated date.
• Pricing is normalized to $/min.
• Providers can vary hardware over time (especially Hetzner-based fleets).

Observations

Performance takeaways

• Shipfox, Namespace, Blacksmith, and RunsOn lead x64 single‑thread performance.
• Cirrus and Namespace lead arm64; Shipfox, RunsOn, and Warpbuild follow. GitHub’s newer Cobalt CPUs are closer but not consistently available.
• GitHub uses older CPUs overall, especially on x64.

Reliability & Scaling

• GitHub queue times are very good for standard runners. Used to be minutes for larger ones, but GitHub has improved their queue system.
• AWS provides better scalability than Hetzner-based providers.
• Third-party SaaS providers may require manual requests for higher vCPU concurrency, sometimes with extra fees. Same for static egress IPs. Self-hosted runners like RunsOn and AWS CodeBuild are not affected by this.
• Variable CPU models on some providers (for example Ubicloud tiers) can impact reproducibility.

Security & infra notes

• GitHub Actions runners are managed by GitHub and run on Azure. Access to private resources often requires secrets or OIDC.
• RunsOn is fully self-hosted in your AWS account with a dedicated GitHub App; IAM policies and networking are fully under your control.
• AWS CodeBuild is managed by AWS, runs in your account, and uses a shared GitHub App for registration.
• Namespace, Blacksmith, Cirrus, Warpbuild, Ubicloud, Depot, Shipfox are SaaS providers; some offer BYOC, but the control plane and runner registration are not fully under your control.

Gaps

• Detailed concurrency/scaling tests are still missing; they matter for high-volume CI/CD (thousands of jobs/day). RunsOn routinely handles more than 1M jobs every day.

Note: this analysis is valid as of February 2026. Things change quickly in this space, so make sure you do your own research as well.