Railway bills hosting by the vCPU-minute, not a named tier

The workload

Railway prices hosting by metered resource consumption rather than a named server size, so the workload before deploying anything is arithmetic: converting expected vCPU and RAM usage into a monthly figure using Railway's own published per-unit rates. Railway's pricing page and pricing reference documentation, both retrieved 16 September 2026, state those rates directly.

What the documents show

Railway's documentation states, verified, that standard container compute is billed at '$20 / vCPU / month ($0.000463 / vCPU / minute)' and RAM at '$10 / GB / month ($0.000231 / GB / minute),' with network egress at '$0.05 / GB.' It states, verified, that a separate, higher-priced tier applies to VM-based workloads such as sandboxes and cloud agents, billed at '$50 / vCPU / month' and '$50 / GB / month.' The pricing page states, verified, that the Hobby plan costs $5 per month and includes '$5 of resource usage per month' toward those metered rates, and the Pro plan costs $20 per month per workspace with '$20 of resource usage per month' included; a free trial provides a one-time $5 credit over 30 days rather than an ongoing free allowance. Both documents state, verified, that billing is metered per second, so a stopped service accrues no compute charge while idle — a structurally different model from a flat named-tier plan that charges the same amount whether a service runs continuously or not.

The operating cost

A workload running 1 vCPU and 1 GB of RAM continuously for a full month would cost, by Railway's own per-minute rates, roughly $20 for the vCPU plus $10 for the RAM, or $30 total before egress — an editorial estimate applying Railway's published rates to an assumed, constant workload, not a Railway-stated bill. On the Hobby plan, the included $5 of usage would cover a small fraction of that before overage billing begins at the same per-unit rates. Railway's own documentation does not state whether metered billing costs more or less than a comparable flat-tier plan for any specific workload, and this entry does not calculate that comparison without both vendors' rates applied to the same named workload.

The stop condition

Railway's documents state the operational stop condition implicitly: because billing is per-second and tied to actual consumption, the only way to stop the meter is to stop the service, not to downgrade a plan tier. The editorial stop condition is a monitoring one: since cost scales continuously with usage rather than in discrete steps, a workload's actual vCPU and RAM consumption should be checked against the included-usage allowance regularly, because a small, gradual increase in resource use will not trigger a plan-tier warning the way crossing a fixed-tier ceiling would.

• Does this workload's actual measured vCPU and RAM usage exceed the plan's included monthly allowance yet?
• Would idle-time auto-stop meaningfully reduce this month's bill, given per-second billing?
• Has egress at $0.05/GB been checked against this workload's actual outbound traffic?

Railway's per-second, per-resource pricing is a documented billing mechanism, not a claim about cost relative to any named competitor; this entry states only what Railway's own pages say about its own rates.