machine.dev vs
Blacksmith.
Blacksmith built a great CI runner. We built a great CI runner with GPU and a spot tier. Same per-vCPU on-demand 26% cheaper. Plus a spot tier 9-12× cheaper.
The overlap.
Blacksmith (YC W24) is the brand leader in this category. 1,000+ orgs, 20M+ jobs/month. Their customer roster includes Vercel, Supabase, Clerk, Mercury, Expensify, Plex, Descript, and Gitbook. They're SOC2 Type 2, GDPR, and ISO 27001. They run on bare-metal gaming CPUs in US + EU colos. Their persistent NVMe Docker layer cache really is fast.
We're not going to pretend any of that doesn't matter. If you need a SOC2 letter today and a customer list with Vercel on it, that's a real reason to pick Blacksmith.
What we offer that they don't: GPU runners (Blacksmith has none), spot pricing (Blacksmith has none, 9-12× cheaper than their pay-as-you-go), and a flat 26% cheaper on x64 on-demand at every size. Read on.
Same sizes. 26% cheaper.
Per-minute USD. Blacksmith prices verified from blacksmith.sh/pricing April 9, 2026. machine.dev prices pulled from the live API.
[NOTE_OD] Both providers scale linearly per vCPU. Blacksmith is $0.002/vCPU/min on x64. machine.dev's effective rate is ~$0.00149/vCPU/min. The 26% gap holds at every size.
[NOTE_SPOT] Blacksmith has no spot tier. The 12.5× spot delta is the structural gap between “managed CI on bare metal” and “managed CI on AWS EC2 with spot.”
ARM: closer race.
Blacksmith's ARM pricing is more aggressive than x64 ($0.00125/vCPU/min). On-demand is roughly tied. Spot still wins decisively.
Beyond price.
| SPEC | BLACKSMITH | MACHINE_DEV |
|---|---|---|
| GPU runners | No | Yes (T4G, T4, L4, A10G, L40S, RTX 6000) |
| AWS Trainium / Inferentia2 | No | Yes |
| ARM GPU (T4G) | No | Yes (only provider in market) |
| Spot tier | No | Yes (8.9-12.5× cheaper than OD) |
| CUDA pre-installed | N/A (no GPU) | Yes (12.1 + cuDNN 9.2.1) |
| Free tier | 3,000 min/month | $10 free compute (one-time) |
| Compliance | SOC2 Type 2, GDPR, ISO 27001 | In progress |
| macOS / Windows | Public beta (M4 Mac, Win x64) | No (Linux only) |
| Persistent NVMe Docker cache | Yes | No (use actions/cache) |
| Customer roster | Vercel, Supabase, Clerk, Mercury, Expensify | YC startups, ML teams, indie devs |
| Free OSS tier | Yes | Same $10 free compute |
| Setup | One runs-on change | One runs-on change |
Five strengths.
Strongest customer roster in the market
Vercel, Supabase, Clerk, Mercury, Ashby, VEED, Plex, Expensify, Chroma, Descript, Gitbook, Finch, Exa. 1,000+ orgs, 20M+ jobs/month. If third-party validation closes deals at your company, Blacksmith has more of it.
SOC2 Type 2, GDPR, ISO 27001
Blacksmith ships compliance documentation today. They're pursuing HIPAA. machine.dev's compliance posture is in progress — if you need a SOC2 letter today, that's a real difference.
Persistent NVMe Docker layer cache
Blacksmith claims 40× faster Docker builds via a persistent NVMe layer cache that survives across jobs. machine.dev relies on actions/cache and doesn't have an equivalent persistent volume story.
macOS + Windows public beta
Blacksmith now offers Apple Silicon M4 (6 / 12 vCPU) and Windows x64 runners in public beta. machine.dev is Linux x64 + ARM64 only.
Larger free tier
3,000 free minutes per month, every month — the most generous in the market. machine.dev offers a one-time $10 free trial.
Five strengths.
26% cheaper on every x64 on-demand size
Both providers scale linearly per vCPU. Blacksmith is $0.002/vCPU/min on x64 on-demand; machine.dev's effective rate works out to ~$0.00149/vCPU. The 26% gap holds at every size from 2 to 32 vCPU.
Spot tier — 9-12× cheaper
Blacksmith has no spot tier at all. machine.dev's spot CPUs are 8.9-12.5× cheaper than Blacksmith's pay-as-you-go on both x64 and ARM. If your jobs tolerate occasional retries, the savings are unambiguous.
GPU runners — they have none
Blacksmith is CPU-only. machine.dev offers six NVIDIA tiers (T4G ARM, T4, L4, A10G, L40S, RTX 6000) plus AWS Trainium and Inferentia2. If your CI pipeline touches model evaluation, training, or any CUDA workload, machine.dev is the only option between the two.
The only ARM GPU runner anywhere
T4G ARM 8 vCPU spot is $0.00277/min ($0.166/hr) — the cheapest GPU runner on the public internet. Built on Graviton g5g.2xlarge. Nobody else offers ARM GPU at any price.
CPU + GPU on one bill
If your team runs both fast CPU CI and GPU evals or training, consolidating on machine.dev means one credit balance, one integration, one runs-on syntax. With Blacksmith you'd need a separate GPU provider.
Same pattern.
More options.
Both providers work the same way: install a GitHub App, change your runs-on label, push. No new tooling, no infrastructure.
Live per-minute rates
Live per-minute rates for all runner types. Transparent pricing with no hidden fees.
Get started free.
$10 free compute on signup. Run your actual workloads and compare before committing.