Cloud Run Cost Calculator

Request-based billing charges CPU and memory only while an instance starts, handles at least one request or shuts down, at $0.000024 per vCPU-second in Tier 1 regions such as us-central1, and adds $0.40 per million requests. Instance-based billing charges for every second the instance exists, busy or not, at $0.000018 per vCPU-second with a 1-minute minimum, and has no per-request fee. So in Tier 1 request-based CPU costs a third more per active second (about 56% more in Tier 2) but nothing for idle time unless you keep minimum instances, and instance-based is the reverse. Which is cheaper depends on what share of instance lifetime is spent actually serving.

Roughly: a busy, steadily loaded service is cheaper on instance-based, because it pays the lower rate on time it was going to use anyway and has little idle to waste. A service whose traffic comes in bursts with long quiet gaps is cheaper on request-based, because instance-based would bill for a lot of time doing nothing. Watch the middle case: Cloud Run keeps an idle instance for up to 15 minutes after its last request, so light but steady traffic keeps an instance alive all day, and instance-based bills every second of it. The crossover is a utilization figure rather than a traffic level, and this calculator reports it for your own inputs.

Concurrency divides the work rather than multiplying it. At the default of 80, eighty simultaneous requests share one instance's CPU and memory allocation, so the vCPU-seconds you pay for are never more than the request-seconds and fall toward request-seconds divided by concurrency as more requests overlap. Light traffic rarely overlaps, so there concurrency changes little. On a busy service, dropping concurrency to 1, which some frameworks require, multiplies the compute bill by up to 80, which makes it the single largest lever on a Cloud Run bill, and it is a one-line configuration change.

Yes, and it is the most common surprise. A minimum instance is held warm to avoid cold starts, and it bills continuously. On request-based billing its CPU is throttled and bills at $0.0000025 per vCPU-second, a tenth of the active rate, but its memory bills at the full rate regardless. On a low-traffic service, min-instances is frequently the largest line on the bill.

One allowance a month per billing account, aggregated across its projects and valued at Tier 1 (us-central1) prices. In seconds it differs by model: 240,000 vCPU-seconds and 450,000 GiB-seconds on instance-based, 180,000 vCPU-seconds, 360,000 GiB-seconds and two million requests on request-based. In dollars both are worth about $4.32 of CPU and $0.90 of memory. It is per billing account rather than per service, so a second service shares it.