Every number, with the method and the date attached
A figure here carries the date it was produced and the method that produced it.
Measured 2026-09-13 through the path a customer uses: the promigence CLI, over the network, an 8 vCPU / 8 GiB sandbox, from clients in three regions.
Measured
Ours, on real repositories.
cold start of an empty sandbox
warm fork on the box
first copy on a server that has never held it
how far one provider's burst moved in 12 days
end to end, through the CLI
forks in six hours, zero failures
at 200 concurrent, all served
sooner on the same CI job, same size
CI runners against hosted CI
The same CI job on our runners, on two hosted CI services and on a self-hosted VM, all at the same size.
| 4 vCPU / 16 GiB | Promigence | Hosted CI A | Hosted CI B | Self-hosted |
|---|---|---|---|---|
| Job start | 1.3–1.6 s | 1–5 s | 0.11–0.31 s | — |
| Install dependencies | 8.8–10.5 s | 20.8–26.7 s | 31.3–32.8 s | 23.2–35.8 s |
| Type-check | 33.6–38.5 s | 60.9–109.8 s | 69.0–91.5 s | 38.9–39.6 s |
| Whole job | 48.5–55.9 s | 105.9–148.7 s | 129.7–182.0 s | 81.8–94.7 s |
| Their default runner | — | never finishes (out of memory) | never finishes (out of memory) | — |
The same unmodified public CI script on every arm: clone a TypeScript monorepo, install, type-check, at 4 vCPU and 16 GiB, n of at least 3 per row, measured 2026-10-01 to 2026-10-03; the self-hosted column is one fresh 4 vCPU / 16 GiB VM, two runs. Our 4 vCPU type-check matched hosted CI A's 8-core runner. Hosted CI B starts a job faster than we do; everything after the start is ours.
Your workload, not ours
Every figure on this page came off a workload we chose, which is the weakest thing about it. The fix is to run yours. Send us the environments and we will run them free, and you get the numbers back whether or not they flatter us.
Run your own workload on it, free
Send a repo and the command you run against it, whatever that is: an eval suite, an RL rollout, a CI job, a queue of coding tasks. We build the environment once, run it a thousand times, and send back the timings, the failures and an exact price. Free, once, on your real workload.
- 1,000 runs of your own command, on your own repo
- What each one cost in wall clock, and anything that failed
- Whether a failure was your code or the environment
- An exact price for your real volume
Not ready to hand over a repo? Read the quickstart or check the numbers first.