A neutral runner guide
Cheaper, faster compute for your CI and your coding agents, measured honestly.
Your CI minutes and your agents' test runs don't have to burn GitHub's prices or your laptop's fans. We're measuring the runners you can use today on price, billing rules, start time and real wall time, and we'll publish every number. Until then, the benchmark shows list prices.
We run Sengi, one of the runners compared here.
Benchmark
GitHub-hosted, Sengi, Ubicloud, AWS on-demand and spot, and Hetzner, at 2, 4 and 8 vCPU. List prices now, measured runs next. Raw data and the ranking rule are public.
Coding agents
Send tests and builds off your laptop with Crabbox, the open-source runner CLI for agents. Sengi as a Crabbox provider is in progress.
CI
Keep GitHub Actions. Most hosted runners switch with one runs-on line per job. See how.
Why per job, not per minute
A cheaper minute is not a cheaper job. A faster CPU finishes in fewer billed minutes, a slow start keeps you waiting, and billing rules decide what a short job really costs: Hetzner bills a started hour, AWS bills boot time by the second, most CI runners round each job up to a minute. So we'll rank by cost per job from measured wall time, and show queue time beside it.
One label, always the cheapest runner
We're considering a single runs-on label that sends each job to whichever runner is cheapest for it right now. Join the waitlist if you'd use it. We'll email you once when there's something to try, and never share your address.
We store only your email address and the date you joined. To be removed, email ops@sengi.run from that address.