Cloud Run
7.4/10Google Cloud's fully managed serverless platform for running containers, functions, batch jobs, and GPU workloads, scaling automatically and billed by usage.
Data last updated November 2026 · Our methodology
Overview
Cloud Run is Google Cloud's managed platform for running containerised code without managing servers or clusters. It offers four ways to run code. Services are stateless HTTP endpoints that autoscale up to about 1,000 instances by default and can scale to zero when idle. Jobs run tasks to completion, either on demand or on a schedule. Worker pools are always-on processes for non-HTTP background work such as consuming Pub/Sub or Kafka queues. Cloud Run functions respond to events. Every service gets an HTTPS endpoint on run.app, and you can add custom domains with managed TLS. You can deploy a prebuilt container image or let source-based deployment build the image for languages such as Go, Node.js, Python, and Java. Code hosted on GitHub can be deployed automatically on each new commit. Services and jobs can attach NVIDIA L4 GPUs for AI inference and training. Pricing is pay-as-you-go per vCPU-second, GiB-second, and request, rounded up to 100 ms. You can choose request-based billing, which charges only while requests are being handled, or instance-based billing, which charges for the instance's whole lifetime. A monthly free tier applies, and flexible committed use discounts are available. It fits naturally with Cloud SQL, Pub/Sub, Eventarc, Cloud Storage, and Firebase. The main drawback is cold-start latency when services scale up from zero. Keeping minimum instances warm avoids this but adds a constant cost. Egress, Cloud Build, and Artifact Registry are also billed separately.
Our Rating
Who Cloud Run is best for
Best software for DevelopersGood fit
Any container or language runs with no cluster management, and you can deploy from source or GitHub.
Autoscaling HTTP services, jobs, and worker pools cover typical SaaS backend needs.
Scale-to-zero and a monthly free tier keep costs low for workloads with little traffic.
It is backed by a 99.95% SLA in most regions and integrates with Google Cloud IAM and its services, though setup needs Google Cloud expertise.
Less suited
Reviewers note that it requires a Google Cloud project, billing setup, and familiarity with IAM, which is too much for non-technical teams.
Pros & Cons
Pros
- Scale-to-zero with request-based billing means idle services cost nothing beyond any minimum instances you set.
- It runs any container image, so there is no language or framework lock-in, and source deploys build images automatically.
- Services, jobs, worker pools, and functions on one platform cover HTTP APIs, batch work, and queue consumers alike.
- Every service gets managed HTTPS on run.app plus custom domains, with WebSockets, HTTP/2, and gRPC support.
- On-demand NVIDIA L4 GPUs billed per second make it practical to serve AI inference without managing GPU VMs.
- A 99.95% monthly uptime SLA applies for non-GPU services in most regions.
Cons
- Cold starts when scaling from zero are the most common complaint, and keeping minimum instances warm adds ongoing cost.
- Internet egress beyond 1 GiB, plus Cloud Build and Artifact Registry, are billed separately and can inflate bills.
- Setup requires a Google Cloud project, a billing account, and familiarity with IAM.
- GPU services must use instance-based billing, get no free tier, and are excluded from committed use discounts.
- Usage-based billing can spike from traffic surges or runaway loops unless you set budgets and alerts.
Key Features
Cloud Run Pricing
Free tier
- No free GPU usage
- 1 GiB of North America egress per month
- Instance-based: 240,000 vCPU-seconds and 450,000 GiB-seconds per month
- Applies per billing account; reported as limited to us-central1, us-east1, and us-west1
- Request-based: 180,000 vCPU-seconds, 360,000 GiB-seconds, and 2 million requests per month
Pay-as-you-go
- Egress billed separately
- NVIDIA L4 GPUs billed per second
- Choice of request-based or instance-based billing
- Billed per vCPU-second, GiB-second, and request, rounded up to 100 ms
- Compute flexible committed use discounts: 17% for request-based services; 28% (1 year) or 46% (3 years) for instance-based services, jobs, and worker pools
Frequently Asked Questions
Yes. Each billing account gets a monthly allowance. On request-based billing this is 180,000 vCPU-seconds, 360,000 GiB-seconds, and 2 million requests. On instance-based billing it is 240,000 vCPU-seconds and 450,000 GiB-seconds. You also get 1 GiB of North America egress. GPUs have no free allowance.
Request-based billing charges CPU and memory only while requests are being processed, plus $0.40 per million requests. Instance-based billing charges for the instance's whole lifetime at lower per-second rates and has no request fee. This suits steady or background workloads.
The Cloud Run SLA targets 99.95% monthly uptime for non-GPU services in most regions. Mexico and Stockholm are lower at 99.9%, and GPU services without zonal redundancy are at 99.5%. Service credits of 10%, 25%, or 50% apply depending on the shortfall.
Set a minimum number of instances so some stay warm. Idle minimum instances on request-based billing are charged at a reduced CPU rate, but they are still an ongoing cost.
Yes. Source-based deployment builds the container image for you, and Cloud Run can deploy new commits from a GitHub repository automatically.
Yes, through Compute flexible committed use discounts. These are 17% for request-based services, and 28% for one year or 46% for three years on instance-based services, jobs, and worker pools. Google no longer sells new standalone Cloud Run commitments.
Similar Tools
See all alternativesAWS Amplify is a pay-as-you-go platform from Amazon Web Services for hosting web apps from Git and building full-stack backends on AWS.
Amazon CloudFront is the AWS content delivery network. It caches and serves websites, APIs and media from more than 750 edge locations, and includes DDoS protection, WAF and edge compute.
Northflank is a Kubernetes-based developer platform for deploying services, jobs, databases and GPU workloads on its managed cloud or on customers' own cloud accounts.
Ranked near Cloud Run in Hosting & Infrastructure
See Cloud Run in the listCloud Run is #13 in Hosting & Infrastructure by our score.
