What GitLab CI costs: hosted runners vs your own runners on Hetzner

• Prices are net, without VAT, as of 19 October 2026. Hetzner's prices are its own USD and EUR list prices. GitLab's and Manage Runners' USD prices are converted to EUR at the ECB reference rate of 16 October 2026 (1 EUR = 1.1253 USD) and marked as converted.
• Each GitLab size is matched to the cheapest Hetzner type in each class with at least the runner's memory, or 2 or 4 times it with a larger server size; x86 only. Cost-Optimized and Regular Performance servers have shared vCPUs, General Purpose servers dedicated ones.
• Jobs are assumed to take as long on a Hetzner server as on GitLab's runner; no benchmark has been run.
• Standard projects only. GitLab's lower cost factors for public projects in GitLab for Open Source, their public forks and community contributions are not included.
• Each server runs one job at a time unless you change it. Running several jobs on a larger server is common, and the calculator supports it with Server size and Jobs per server; the jobs then share the server's CPU and RAM.
• Hetzner bills each server by the started hour, up to its monthly price. An always-on server costs its monthly price. A schedule deletes the server after every window, so each window is billed on its own, rounded up to whole hours; the monthly cap never applies across windows. The few minutes a server takes to start are ignored, and a month averages 730 hours.
• With autoscaling, the anchor, sized for the baseline jobs in parallel, always runs or runs on its schedule. Burst runners of the same server type cover the rest of the peak, at the same time on every peak day and only while the anchor runs. Autoscaling works with GitLab runners only.
• Each burst runner is a new server for every peak window. The calculator counts the peak hours plus 0.25 h, rounded up to whole hours and up to the monthly price. The 0.25 h is an allowance for the few minutes to start and for scale-in, which starts draining a burst runner near the end of a paid hour. Hetzner bills started hours, so draining a long job into a further hour costs that hour too; this is not counted.
• Every runner keeps its primary IPv4 address ($0.60 / €0.50 a month), also while paused. Burst runners are deleted with their addresses, so they pay IPv4 only for their billed hours.
• The Manage Runners plan is counted once per organisation: Basic when runners are always on, Pro when they use a schedule or autoscaling. Choose "I run GitLab Runner myself" to leave it out.
• GitLab seat prices are not counted, because your own runners don't change them. GitLab's minutes are counted at the steady-state price; packs of 1,000 carry over.
• Not counted: Hetzner traffic beyond the included volume, a cache bucket in object storage, and your own time.

GitLab
  compute minutes  = job minutes × cost factor
  extra minutes    = max(0, compute minutes − quota)
  GitLab per month = extra minutes ÷ 1,000 × price per 1,000

Hetzner, per server
  server type      = cheapest in its class with RAM ≥ runner RAM × server size
  always on        = monthly price
  on a schedule    = windows × min(ceil(hours per day) × hourly price, monthly price)
  windows          = days per week × 52 ÷ 12
  per month        = servers × (server + IPv4)
  servers          = ceil(parallel jobs ÷ jobs per server)
  running hours    = 730 always on, hours per day × windows on a schedule

With autoscaling, per class
  anchor servers   = ceil(baseline jobs ÷ jobs per server)
  burst servers    = ceil(parallel jobs ÷ jobs per server) − anchor servers
  billed hours     = ceil(peak hours per day + 0.25)
  peak windows     = peak days per week × 52 ÷ 12
  burst server     = peak windows × min(billed hours × hourly price, monthly price)
  burst IPv4       = IPv4 per month ÷ 730 × billed hours × peak windows
  per month        = anchor servers × (server + IPv4)
                     + burst servers × (burst server + burst IPv4)

Your own runners   = Hetzner per month + Manage Runners plan
Break-even         = (quota + own runners ÷ price per compute minute) ÷ cost factor   [job minutes]
Job minutes the quota covers = quota ÷ cost factor
Job slots busy     = job minutes ÷ capacity minutes
  capacity minutes = servers × running hours × jobs per server × 60
  with autoscaling = (anchor servers × running hours + burst servers × peak hours per day × peak windows)
                     × jobs per server × 60