An open, static dataset of CI runner prices, specs and billing rules: what a runs-on: label costs per minute, what hardware it gets, and how the provider bills it. It works like models.dev does for LLMs, but for CI runners.
- Browse: https://runnerdb.stamp.build
- API: https://runnerdb.stamp.build/api.json
- Schema: https://runnerdb.stamp.build/schema.json
- Cost calculator: https://runnerdb.stamp.build/calculator/ (monthly cost of every runner, per-minute and flat-fee, for your minutes and concurrency; runs in the browser)
Every price has a source_url and a verified_as_of date, so you can see where a number came from and when someone last checked it. Hardware specs (vCPU, RAM, disk) come from the same page or, when the provider publishes them elsewhere, from the runner's specs_url.
| Provider | Directory | Runners |
|---|---|---|
| GitHub-hosted | providers/github |
Linux x64/ARM, Windows, macOS standard + macOS large/xlarge |
| Blacksmith | providers/blacksmith |
Linux x64/ARM 2-32 vCPU, Windows 4-16 vCPU, macOS 6/12 vCPU |
| Depot | providers/depot |
Linux x64/ARM 2-64 vCPU, Windows 2-64 vCPU, macOS 8 vCPU |
| Namespace | providers/namespace |
Linux x64/ARM, Windows and macOS standard shapes (unit-minute billing) |
| Ubicloud | providers/ubicloud |
Linux x64 standard + premium and ARM, 2-30 vCPU |
| WarpBuild | providers/warpbuild |
Linux x64/ARM 2-32 vCPU, ARM on Apple Silicon, Windows 4-32 vCPU, macOS 6/12 vCPU |
| RunsOn | providers/runs-on |
Flex predefined Linux runners. $0 per minute: you pay AWS directly plus a flat annual license |
| Cirrus Runners | providers/cirrus-runners |
macOS, Linux x64/ARM, Linux GPU. $0 per minute: flat $150/month per concurrency unit |
Providers that bill a flat fee or license instead of per minute (RunsOn, Cirrus Runners) have billing.unit: "flat", price_usd_per_minute: 0 and a flat_fee on each runner. They aren't directly comparable with per-minute prices, so the site lists them on a separate page, Flat-fee & license plans, not in the main table. The cost calculator puts both kinds side by side: per-minute runners cost minutes × price_usd_per_minute, flat-fee runners flat_fee.amount_usd (per month, or per year / 12) × concurrent jobs. Runners without a fixed amount_usd (RunsOn: AWS cost + license) are listed but not ranked. api.json includes all providers. Each provider's billing.notes describes its real cost model.
Wanted, contributions welcome: GitLab SaaS runners, CircleCI, GitHub Linux/Windows larger runners and GPU runners. Please add only prices you can link to a public source. (BuildJet for GitHub Actions has shut down and is not listed.)
providers/
<provider-id>/
provider.json # provider + billing rules
runners/
<runner-id>.json # one file per runner type
schema/
provider.schema.json # JSON Schema (draft 2020-12)
runner.schema.json
build.mjs # validator + api.json/schema.json builder (no dependencies)
scripts/fetch-prices.mjs # daily price check against the providers' pricing pages (no dependencies)
site/ # the web page: Vite + React + Tailwind, shadcn/ui + TanStack Table
| Field | Required | Description |
|---|---|---|
name |
yes | Display name |
homepage |
yes | https URL |
pricing_url |
yes | Official pricing page |
billing.unit |
yes | minute, second or flat (flat fee or license, not metered per minute: every runner has flat_fee and price_usd_per_minute: 0) |
billing.rounding |
yes | per_job_ceil (each job rounded up to a whole unit), per_second, none, unknown |
billing.base_rate_usd |
no | For multiplier-based billing: USD per billing minute at 1x. Every runner must then set multiplier, and price_usd_per_minute must equal base_rate_usd × multiplier (the build checks this). |
billing.notes |
no | Free text |
public_repo_policy |
yes | standard_runners_free, billed, not_applicable |
free_tier |
no | { notes, minutes_per_month? }. Informational only; never applied to prices. |
notes |
no | Free text |
| Field | Required | Description |
|---|---|---|
id |
yes | Must equal the file name (without .json) |
labels |
yes | Exact runs-on labels. * matches one or more characters (e.g. macos-*-large). A label can belong to only one runner across the whole dataset. |
os |
yes | linux, windows, macos |
os_version |
no | Image hint, e.g. ubuntu-24.04 |
arch |
yes | x64, arm64 |
vcpu |
yes | Integer, or null if not verified |
memory_gb |
no | Memory (RAM) in GB, as published by the provider |
disk_gb |
no | Disk (SSD/storage) the runner gets, in GB, as published by the provider. If the provider states something more nuanced (e.g. "14 GB free", "1.5 TB"), store the number in GB and explain in notes |
gpu |
no | { model, count, memory_gb? } |
price_usd_per_minute |
yes | List price per runtime minute, before plan quotas, discounts and rounding |
flat_fee |
with billing.unit: "flat" |
{ summary, amount_usd?, period?, notes? }: summary is short display text (e.g. $150/mo per concurrent job); amount_usd is the fee per period (month or year) when one fixed amount applies. Required for, and only allowed on, flat-billed providers. |
multiplier |
no | Billing-minute multiplier (required when the provider sets base_rate_usd) |
public_repo_free |
yes | true if jobs from public repositories are not billed |
scheduling_discounts |
no | Price factors for optional modes, e.g. {"flex": 0.6} = 60% of list price |
source_url |
yes | https URL that backs the price |
specs_url |
no | https URL of the official spec page that backs vcpu, memory_gb, disk_gb and gpu, when that is a different page from source_url (e.g. GitHub's runner reference vs. its pricing reference). If omitted, source_url backs the specs too |
verified_as_of |
yes | YYYY-MM-DD when the price was last checked against source_url |
notes |
no | Free text |
curl -s https://runnerdb.stamp.build/api.json | jq '.providers.github.runners["linux-2-core-x64"].price_usd_per_minute'api.json shape:
To price a job: look up the runner whose labels match the job's runs-on, then compute ceil(minutes) × price_usd_per_minute when billing.rounding is per_job_ceil. Multiply by scheduling_discounts.<mode> if the job uses that mode, and check public_repo_free for public repositories. Plan quotas and negotiated discounts are out of scope.
Prices change. Use as_of/verified_as_of to decide how much to trust a number, and confirm against source_url before relying on it for anything that matters.
A scheduled workflow (.github/workflows/daily-prices.yml, daily at 06:17 UTC, or run it by hand from the Actions tab) runs scripts/fetch-prices.mjs. The script has one fetcher per provider. Each fetcher downloads the provider's official pricing source (a markdown or JSON version where the provider publishes one, otherwise the pricing page HTML) and reads every runner's price:
- Price changed: writes the new
price_usd_per_minute(andmultiplierfor multiplier-billed providers) and setsverified_as_ofto today. - Price unchanged: sets
verified_as_ofto today. - Runner or price not found (usually a page redesign): that provider's files are left untouched and the script exits non-zero. A stale price is never re-dated.
- SKU on the pricing page but not in runnerdb: listed in the summary, never added automatically.
The workflow then runs node build.mjs --check and opens or updates a pull request from bot/daily-prices, with the markdown summary as its body:
- Price changes are proposed the same day ("Update runner prices (N changed)").
- Date-only re-verifications are committed weekly, on Mondays or when the workflow is run by hand ("Re-verify prices (no changes)"). A daily PR that touches more than 100 files just to move a date would be noise, and a
verified_as_ofup to a week old is still recent enough. While a bot PR is open, it is refreshed every day. If the data matches the default branch again, the PR is closed. - If any fetcher fails, the job still opens the PR for the providers that checked out, then fails, so GitHub notifies the maintainers.
Run it locally:
node scripts/fetch-prices.mjs --dry-run # check everything, write nothing
node scripts/fetch-prices.mjs --provider depot # check one provider and write the resultThe workflow uses GITHUB_TOKEN and needs the repository setting Allow GitHub Actions to create and approve pull requests (Settings > Actions > General). PRs opened with GITHUB_TOKEN don't trigger other workflows, so check.yml won't run on them; the price workflow runs node build.mjs --check itself before it pushes.
Short version: one PR per price change, with source_url and an updated verified_as_of. See CONTRIBUTING.md.
node build.mjs --check # validate only (CI runs this)
node build.mjs # validate + write dist/api.json, dist/schema.jsonFor data changes you need Node 18 or newer and nothing else; there is no npm install.
The web page lives in site/ (Vite, React, Tailwind CSS, shadcn/ui components and TanStack Table). It embeds dist/api.json at build time and writes dist/index.html and dist/assets/ next to the JSON files. It needs Node 20.19 or newer:
npm ci --prefix site
npm run build --prefix site # runs node build.mjs, then vite build into dist/
npm run dev --prefix site # local dev serverAmplify runs the same two commands (see amplify.yml).
-
Data (
providers/,schema/, generatedapi.json/schema.json): CC0-1.0. Public domain; use it for anything. -
Code (
build.mjs,site/, the generated page, workflow): MIT.site/src/components/ui/is adapted from shadcn/ui (MIT). -
Provider names and logos: trademarks of their respective owners, not covered by MIT or CC0. runnerdb is not affiliated with or endorsed by any provider; logos are used only to identify each provider. See NOTICE.
{ "$schema": "https://runnerdb.stamp.build/schema.json", "as_of": "2026-11-18", // newest verified_as_of in the dataset "providers": { "<provider-id>": { "id": "...", "name": "...", "billing": { ... }, ..., // provider.json fields "as_of": "2026-11-18", // newest verified_as_of among its runners "runners": { "<runner-id>": { ..., // runner file fields "provider": "<provider-id>", "as_of": "2026-11-18" // = verified_as_of } } } } }