AWS CodeBuild pricing in 2026
./codebuild --per-build-minute --no-platform-fee
How much does AWS CodeBuild cost in 2026?
CodeBuild is pure pay-as-you-go: no platform fee, no per-seat charge, you pay only for build time. On-demand Linux compute in US East starts at $0.005 per build minute on general1.small, rising to $0.010 (medium) and $0.020 (large). ARM (Graviton) is cheaper again, from $0.0034 per minute. The AWS Free Tier includes 100 build minutes a month. Builds are billed by the minute (per second on Lambda compute), so the total is simply your compute rate times your monthly build minutes.
AWS CodeBuild is Amazon's fully managed build service. Unlike the seat-plus-minute platforms, it has no plans and no users to license: you point it at a source repo, pick a compute size, and pay per build minute. That makes it one of the cheapest managed CI options at volume, and the natural pick for teams already deep in AWS. The prices below are the on-demand rates from the public AWS Price List API (published 4 August 2026, US East / N. Virginia) and the CodeBuild pricing page. Rates vary by region; check the live figures for yours before committing.
On-demand compute rates (Linux, US East)
Every build runs on a compute type you choose. Bigger machines cost more per minute but finish faster, so the cheapest total often comes from matching the size to the workload rather than always picking the smallest. ARM (Graviton) machines are the cheapest at every size if your toolchain runs on ARM.
| Compute type | vCPU / memory | $ / build minute |
|---|---|---|
| arm1.small | 2 vCPU / 3 GB | $0.0034 |
| general1.small | 2 vCPU / 3 GB | $0.005 |
| arm1.medium | 4 vCPU / 8 GB | $0.007 |
| general1.medium | 4 vCPU / 7 GB | $0.010 |
| arm1.large | 8 vCPU / 16 GB | $0.015 |
| general1.large | 8 vCPU / 15 GB | $0.020 |
| general1.xlarge | 36 vCPU / 72 GB | $0.0798 |
| general1.2xlarge | 72 vCPU / 144 GB | $0.20 |
Source: AWS Price List API (CodeBuild offer, publicationDate 4 August 2026), US East (N. Virginia), on-demand EC2 compute. Windows: general1.medium $0.018/min, general1.large $0.036/min. GPU (gpu1.large, 48 vCPU / 192 GB) $0.30/min. Rates vary by region.
The billing model in one paragraph
There is nothing to plan for and no seats to count. You are billed per build minute (rounded up to the minute) on EC2 compute, or per second on Lambda compute, from build start to termination. The first 100 build minutes each month on general1.small or arm1.small are free, and that free tier does not expire after a year. Everything else on the bill (artifact storage in S3, data transfer, any KMS or Secrets Manager calls your build makes) is billed by those services separately, not by CodeBuild. For most pipelines the compute-minute line is the whole story.
Real monthly cost at three team sizes
These are our own models: the on-demand rate above times a realistic monthly build-minute volume for each team. Swap in your own minute count and the arithmetic is the same. All assume Linux on-demand and subtract the 100 free minutes.
4,000 build minutes minus 100 free, times $0.005, is about $19.50. On arm1.small ($0.0034) the same volume is closer to $13.
25,000 minutes on general1.medium at $0.010 is $250. Dropping to general1.small halves it to ~$125 if 2 vCPU is enough; arm1.medium ($0.007) lands near $175.
100,000 minutes at $0.010 is $1,000. On arm1.medium ($0.007) it is ~$700. At this volume reserved-capacity fleets or self-hosted runners start to pay off.
Where CodeBuild wins and where it does not
CodeBuild is at its best when your world is already AWS. Pulling base images from ECR, assuming an IAM role to deploy to ECS, EKS or Lambda, reading secrets from Secrets Manager, writing artifacts to S3 - all of it is native, with no cross-cloud egress and no long-lived credentials to manage. The pay-per-minute model with no platform fee means a small or bursty workload pays almost nothing, and the ARM compute makes it one of the cheapest managed meters at scale.
It is a weaker fit when you want a batteries-included CI experience. There is no marketplace of pre-built steps the way GitHub Actions has, no hosted macOS for iOS builds, and the pipeline UX is spartan compared with GitLab CI or CircleCI. Teams typically pair CodeBuild with CodePipeline (or an external orchestrator) rather than treating it as a full CI/CD platform on its own. If your source lives on GitHub and you want the richest ecosystem, GitHub Actions is usually the better home even though its per-minute rate is slightly higher.
Compare with other platforms
Frequently Asked Questions
# click any question to expand