Google Cloud Platform Performance: Benchmarks, Latency & Limits 2026
Cloud Run bills per second, scales on concurrency to 1,000 and to zero, with startup CPU boost on by default. A 99.95% SLA covers all 43 regions.
Google Cloud Platform Performance verdict
Cloud Run runs serverless containers billed by the second, at $0.000024 per vCPU-second and $0.0000025 per GiB-second.
But only while a request runs, plus $0.40 per million requests with 2M free monthly. It scales on concurrency, 80 by default up to 1,000, and down to zero, using default-on startup CPU boost to cut cold starts.
The billing model rewards bursty work and treats idle waiting differently than by-the-hour clouds. For latency-sensitive services, keep min-instances at 1 or more and lean on startup CPU boost. For bursty or internal work, let it scale to zero and pay only per request. Budget by active vCPU-seconds rather than the hour, and use the 2M free requests each month. Cap max-instances to bound the bill. Pick a Tier 1 region for the lower rate.
- Cloud Run scales to zero by default, so an idle service cold-starts on the next request. Set min-instances to keep a warm floor, billed at the idle rate.
- The independent 1,090ms cold start is 2020 data from before startup CPU boost, which is now on by default and lowers it. Request-based billing charges only during request processing, so an always-busy service costs far more than a bursty one.
- Tier 2 regions are priced higher than Tier 1 regions.
- Warm latency (independent)
- 32 ms
- CPU rate (active)
- $0.000024/vCPU-s
- Max concurrency
- 1,000
- SLA
- 99.95%
- Free requests
- 2M/mo
This page covers Cloud Run's latency, scaling and cost. Region coverage lives on its own page.
Estimate your Google Cloud Platform usage cost
- 1 × (1 vCPU + 1 GB) for 730 h/mo costs about $69.64/mo (91% CPU, 9% RAM), running always-on.
- Request-based billing: 2M requests, 180k vCPU-seconds and 360k GiB-seconds are free each month, billed per 100ms only while a request runs. Add $0.40 per million requests.
Estimated from Google Cloud Platform's published per-minute rates ($0.00144/vCPU-min, $0.00015/GB-min). Egress and storage are billed separately. Verify against your own workload.
Latency and reliability
| Metric | Value | Source |
|---|---|---|
| Warm latency (independent) | 32 ms (median) | IOD, 2020 |
| Cold start (independent) | 1,090 ms | IOD, 2020, pre-boost |
| Cloud Run SLA | 99.95% | Cloud Run SLA |
| SLA, Mexico / Stockholm | 99.9% | Cloud Run SLA |
| Billing granularity | Per 100 ms | Cloud Run pricing |
| Global network | 10M km fiber, 200+ edge | GCP locations |
Autoscaling and concurrency
| Capability | Value | Notes |
|---|---|---|
| Concurrency | 80 default, 1,000 max | Max simultaneous requests per instance |
| Scale to zero | Default | Instances terminated when there are no requests |
| Max instances | 100 / revision (raisable) | Set at service or revision level |
| Min instances | Warm floor, idle rate | Avoids cold starts; billed at a lower idle rate |
| Startup CPU boost | Up to 2x vCPU | During startup + 10s; on by default |
| Request timeout | 300 s default, 3,600 s max | Configurable per service |
Google Cloud Platform resources and prices
| Resource | Range / rate | Notes |
|---|---|---|
| vCPU per instance | 0.08, 8 vCPU | Default 1 vCPU |
| Memory per instance | up to 32 GiB | 32 GiB requires 8 vCPU |
| CPU (request-based) | $0.000024 / vCPU-sec | Active time only |
| Memory (request-based) | $0.0000025 / GiB-sec | Active time only |
| CPU (instance-based) | $0.000018 / vCPU-sec | CPU always allocated |
| Requests | $0.40 / 1M (2M free) | Request-based billing only |
| Free grant | 180k vCPU-s + 360k GiB-s | Per month, request-based |
Google Cloud Platform reliability and architecture
- Cloud Run's non-GPU SLA is 99.95% monthly uptime for all regions except Mexico and Stockholm, where it is 99.9%, with downtime defined as a 5XX error rate above one percent
- Cloud Run GPU services get 99.95% with zonal redundancy and 99.5% without
- Startup CPU boost temporarily raises a container's CPU (a 0-1 vCPU container is boosted to 2 vCPUs) during startup and for ten seconds after, on by default, to cut cold-start latency
- Setting a minimum number of instances keeps containers warm to avoid cold starts, billed at a lower idle rate under request-based billing
- Cloud Run runs on Google's global network: 10 million km of terrestrial and subsea fiber and 200+ network edge locations across 200+ countries
- There is no charge for data transfer from Cloud Run to Cloud CDN, Media CDN or Cloud Load Balancing
Google Cloud Platform latency benchmarks, independently measured
- An independent benchmark (IOD, May 2020) ran a Go service from a VM in the same region (us-central1) and measured Cloud Run's median warm latency at 32ms, about 37% lower than AWS Lambda's 51ms
- The same benchmark measured Cloud Run's cold start at 1,090ms versus Lambda's 210ms, making Lambda about 5.2x faster on cold starts at the time
- That benchmark predates startup CPU boost; treat the 1,090ms cold start as a 2020 upper bound that current boost-enabled services beat
- Cloud Run bills only while a request is processed (or during startup and shutdown) under request-based billing, rounded up to the nearest 100 milliseconds
- Cloud Run is available in all 43 Google Cloud regions, split into 20 lower-priced Tier 1 and 23 Tier 2 pricing regions
Google Cloud Platform Performance FAQ
How is Cloud Run priced?
By the second. Request-based billing, the default, charges $0.000024 per vCPU-second and $0.0000025 per GiB-second only while a request is processed, or during startup and shutdown. It adds $0.40 per million requests, rounded to the nearest 100 ms. Instance-based billing, with CPU always on, is $0.000018 per vCPU-second. A monthly free grant covers 180,000 vCPU-seconds, 360,000 GiB-seconds and 2 million requests.
How does Cloud Run autoscaling and concurrency work?
Cloud Run scales the instance count to the incoming load, using concurrency, the maximum simultaneous requests per instance, as the lever, 80 by default and up to 1,000. It scales to zero when idle and up to 100 instances per revision by default, which you can raise. Set min-instances for a warm floor and max-instances to bound cost. An instance runs up to 8 vCPU and 32 GiB.
Does Cloud Run have cold starts?
Yes, because it scales to zero by default, so the first request after idle starts a new instance. An independent 2020 benchmark measured that cold start at 1,090 ms against a 32 ms warm request. Google has since added startup CPU boost, on by default, which temporarily doubles CPU during startup to cut the delay. Setting min-instances to 1 or more avoids cold starts entirely.
What is Cloud Run's SLA?
99.95% monthly uptime for non-GPU services in every region except Mexico and Stockholm, which get 99.9%, where downtime means a 5XX error rate above one percent. GPU services get 99.95% with zonal redundancy or 99.5% without. It all runs on Google's global network, 10 million km of fiber and 200+ network edge locations.
How does Cloud Run compare to AWS Lambda on performance?
An independent same-region benchmark from IOD in 2020 found Cloud Run about 37% faster than Lambda when warm, 32 ms against 51 ms median. On a cold start it ran roughly 5x slower, 1,090 ms against 210 ms at the time. That cold-start gap predates startup CPU boost, which Google added specifically to close it, so current figures beat the 2020 numbers.
Sources & verification
| Source | What was checked | Last checked |
|---|---|---|
| Google Official | Official product page | July 10, 2026 |
| Google About Locations | About Locations | July 10, 2026 |
| Google Configuring Max Instances | Configuring Max Instances | July 10, 2026 |
| Google Configuring Request Timeout | Configuring Request Timeout | July 10, 2026 |
| Google Run About Concurrency | Run About Concurrency | July 10, 2026 |
| Google Run Locations | Run Locations | July 10, 2026 |
| Google Run Pricing | Pricing and plans | July 10, 2026 |
Every fact on this Google Cloud Platform page is tied to a named source and a verification date. Freshness-sensitive figures trace to the sources above; verify against the vendor before relying on them.
Explore Google Cloud Platform
Every page on Google Cloud Platform in one place, you are on performance.