Blacksmith

What Blacksmith does

Blacksmith provides managed, high-performance runners for GitHub Actions. Teams keep their workflow YAML and GitHub pull-request experience, but jobs execute on Blacksmith infrastructure. The vendor says its hardware is twice as fast as GitHub-hosted runners, cache downloads are four times faster, and queue time is eliminated. Those are vendor benchmarks rather than universal outcomes, yet the product's core distinction is concrete: faster compute, integrated caching, and Actions observability without asking teams to adopt different CI syntax.

Blacksmith complements adjacent developer tools. Harness covers broader delivery and deployment workflows, Applitools specializes in visual testing, and Datadog LLM Observability monitors production AI systems. Blacksmith's narrower job is accelerating the runner layer behind existing Actions. That makes migration relatively straightforward, but workflow design, flaky tests, and inefficient dependency graphs remain the team's responsibility.

Current pricing

The September 2026 page listed Ubuntu x64 at $0.004 per minute, Ubuntu ARM at $0.0025 per minute, Windows x64 at $0.008 per minute, and macOS M4 at $0.08 per minute for displayed runner sizes. Accounts receive 3,000 free minutes per month. Optional Docker layer caching and sticky disks each cost $0.50 per GB per month; static IPs cost $100 per IP per month. Enterprise is quote-based and adds capabilities such as enterprise concurrency and CI optimization support.

The vendor's “up to 67%” Ubuntu savings combines a claimed 2x runtime improvement with a 33% lower per-minute rate. Do not apply that percentage to the entire current bill without measurement. Repository mix, parallelism, cache hit rate, operating system, artifact transfer, and setup time affect the result. Compare completed-workflow cost, not only the minute rate.

Strengths, drawbacks, and testing

Blacksmith's major strengths are GitHub Actions compatibility, simple usage pricing, ARM and Apple Silicon options, and cache acceleration. It is attractive when CI is constrained by CPU, dependency downloads, Docker builds, or runner queues. The free allowance gives small teams a credible no-card pilot.

The main drawback is platform concentration: organizations using GitLab CI, CircleCI, or a mixed estate do not get one universal runner layer. Faster hardware cannot repair serial test architecture or nondeterministic tests. Add-on storage and static IP charges can become material, and an external runner provider adds another availability and security dependency.

Benchmark at least 50 representative workflows over two weeks. Record p50 and p95 queue time, setup time, test duration, cache hit rate, Docker-build time, failure rate, and dollars per successful run. Include cold-cache and warm-cache trials. Verify OIDC behavior, secret isolation, network egress, log retention, artifact handling, SOC reports, region availability, concurrency caps, and incident procedures. A strong result is lower cost per merged pull request with unchanged reliability—not merely one impressive demo build.