Zest
ci cd platforms

CI/CD Platforms Compared: 12 Tools, Pricing, Best For

By Zest | Updated | 15 min read

Related guide: DORA metrics and AI coding

Two monitors next to a server rack, one screen showing a pipeline diagram under the words Top CI/CD Platforms

Use the CI/CD platform that lives next to your code unless you have a reason not to: GitHub Actions if your repositories are on GitHub, GitLab CI/CD if you want source control, pipelines and security scanning in one product, Bitbucket Pipelines on Bitbucket.

Reach for Jenkins when you need full control on your own servers, Buildkite when you want a hosted control plane with builds on your hardware, and your cloud’s own service (Cloud Build, CodePipeline, Azure Pipelines) when nearly everything deploys there.

Below: a pricing table, then each of the 12 platforms in a few paragraphs, and how to choose. Prices change often; every figure here was checked on the vendor’s own pricing page on October 4, 2026, or on November 10 for GitHub and for the runner prices added then.

What a CI/CD platform does

A CI/CD platform, also called a CI/CD system, a CI server or a continuous integration platform, is the software that builds, tests and deploys your code every time it changes. Continuous integration builds the code and runs the tests on every push or pull request. Continuous delivery keeps every passing change ready to release, and continuous deployment goes one step further and releases it automatically. Almost every platform keeps the pipeline as code, in a YAML file, a Jenkinsfile or a Kotlin DSL committed next to the code it builds. CI is only as useful as the tests it runs, which is why red-green-refactor and CI go well together.

Pricing and free tiers at a glance

Platform Free tier Paid from Where builds run
GitHub Actions Public repos free; private $12/month of usage, about 2,000 Linux min (Free plan) Linux $0.006/min Hosted or self-hosted
GitLab CI/CD 400 compute min/month, 5 users Premium $29/user/month (10,000 min) Hosted or self-managed
CircleCI 30,000 credits/month (up to ~6,000 min), 5 users $15/month (5 users) Hosted or self-hosted
Jenkins Free, open source Your servers Self-hosted
Azure Pipelines 1 hosted job, 1,800 min/month; 1 self-hosted job $40/month per extra hosted job Hosted or self-hosted
Bitbucket Pipelines 50 min/month, 5 users Standard $3.65/user/month (2,500 min) Hosted or self-hosted
Buildkite 5 users, 2,000 hosted Linux vCPU-min/month Pro $30/active user/month Your agents or hosted
TeamCity On-prem: 3 agents, 100 build configurations Cloud $15/committer/month Self-hosted or cloud
Harness CI 2,000 cloud credits/month (~2,000 Linux min) Contact sales Hosted or self-hosted
Google Cloud Build 2,500 build-min/month per billing account $0.006/min (e2-standard-2) Google Cloud
AWS CodePipeline + CodeBuild 100 action-min + 100 build-min/month $0.002/action-min; build from $0.005/min AWS
Semaphore $15 credits/month (~2,000 min); open-source edition $0.0075/min (2 vCPU Linux) Hosted or self-hosted

1. GitHub Actions

GitHub Actions runs workflows defined in YAML files under .github/workflows, triggered by pushes, pull requests, schedules or almost any repository event. The Marketplace has thousands of reusable actions, from cloud logins to Slack notifications.

  • Best for: teams and open-source projects already on GitHub.
  • Pricing: free on standard runners for public repositories. Private repositories draw on a monthly allowance that GitHub now states in dollars, $12 on the Free plan (about 2,000 Linux minutes), $18 on Pro and Team (about 3,000) and $300 on Enterprise Cloud (about 50,000), so Windows and larger runners use it up faster (GitHub Actions billing). GitHub cut hosted-runner prices by up to 39% on January 1, 2026; a Linux 2-core minute is now $0.006 and Windows $0.010. Self-hosted runners are free. A $0.002-per-minute fee for self-hosted runners, announced for March 2026, was postponed.
  • Pros: no setup beyond a YAML file; pull request checks, environments and secrets are built in.
  • Cons: macOS minutes cost far more than Linux; YAML spread across many repositories needs reusable workflows to stay manageable.

2. GitLab CI/CD

GitLab CI/CD reads a .gitlab-ci.yml file and sits inside GitLab’s single application, next to merge requests, issues and the container registry. You can use GitLab.com or run GitLab on your own servers.

  • Best for: organizations that want one DevSecOps platform, with source control, CI/CD and security scanning from the same vendor.
  • Pricing: Free has 400 compute minutes a month and 5 users per private group. Premium is $29 per user a month, billed annually, with 10,000 minutes; Ultimate (50,000 minutes) is priced on request (GitLab pricing).
  • Pros: parent-child pipelines, the CI/CD Catalog of reusable components and merge trains keep large setups manageable. Basic SAST and secret detection run on every tier.
  • Cons: DAST, security dashboards and most compliance features need Ultimate. Self-managed GitLab is a sizable system to operate.

3. CircleCI

CircleCI is a hosted CI service known for performance features: test splitting across parallel containers, Docker layer caching, and reusable config packages called orbs. You can SSH into a failed job to debug it, which helps when the failure won’t reproduce in your local debugger.

  • Best for: teams that want to tune build speed on hosted infrastructure, including macOS, Windows, Arm and GPU machines.
  • Pricing: Free includes 30,000 credits a month (up to about 6,000 build minutes on a small Docker machine) and 5 active users. Performance starts at $15 a month for the first 5 users, plus $15 per extra user; extra credits are $15 per 25,000 (CircleCI pricing). Since December 1, 2025, users without a CircleCI account count as active users.
  • Pros: strong parallelism and caching; good insight into slow and flaky tests.
  • Cons: credit pricing takes effort to forecast; macOS and large machines burn credits quickly.

4. Jenkins

Jenkins is the open-source automation server that many teams started with. Pipelines live in a Jenkinsfile in declarative or scripted syntax, and a large plugin ecosystem connects it to nearly any tool.

  • Best for: teams that need full control on their own infrastructure, or deep integration with legacy and in-house systems.
  • Pricing: free. You pay for servers and for the engineers who maintain them.
  • Pros: runs anywhere; no license fees; distributed agents scale out builds.
  • Cons: you own upgrades, security patches, backups and plugin compatibility. The current LTS line (2.580.2, October 2026) needs Java 21 or 25; Java 17 support ended with 2.555.1 in April 2026 (Jenkins LTS changelog).

5. Azure Pipelines

Azure Pipelines is the CI/CD part of Azure DevOps. It builds code from Azure Repos, GitHub or Bitbucket with YAML pipelines, and its multi-stage pipelines support environments with approvals and checks.

  • Best for: Microsoft and .NET shops, and teams already on Azure DevOps Boards and Repos.
  • Pricing: private projects get one Microsoft-hosted job with 1,800 minutes a month (60 minutes per job) and one self-hosted job with unlimited minutes. Extra parallel jobs cost $40 a month hosted or $15 a month self-hosted (Azure DevOps pricing).
  • Pros: first-class Windows and .NET agents; traceability from work item to deployment.
  • Cons: the interface spans several Azure DevOps services and takes time to learn. Public projects are being retired: no new ones since April 2026, and existing ones become private in 2027, so open-source projects lose their unlimited free minutes.

6. Bitbucket Pipelines

Bitbucket Pipelines is built into Bitbucket Cloud. A bitbucket-pipelines.yml file defines steps that each run in a Docker container.

  • Best for: teams on Bitbucket and Jira.
  • Pricing: Free (up to 5 users) has 50 build minutes a month; Standard is $3.65 per user a month with 2,500 minutes, Premium $7.25 with 3,500. Extra minutes are $10 per 1,000 (Bitbucket pricing).
  • Pros: almost no setup for Bitbucket repositories; build and deployment status shows up on Jira issues.
  • Cons: 50 free minutes run out fast; fewer machine types and integrations than dedicated CI services.

7. Buildkite

Buildkite splits CI in two: Buildkite hosts the control plane and web interface, and agents run the builds on machines you choose, from cloud instances to a rack of Mac minis. Pipelines are YAML in your repository, and they can generate more steps dynamically.

  • Best for: large teams that need high concurrency and want code and secrets to stay on their own infrastructure.
  • Pricing: Free covers 5 users, 10 concurrent jobs and 2,000 hosted Linux vCPU-minutes a month. Pro is $30 per active user a month for up to 50 users; hosted Linux agents cost $0.004 per vCPU-minute (Buildkite pricing).
  • Pros: scales to very large monorepos; signed pipelines, SSO and audit logs.
  • Cons: with self-hosted agents you run and scale the build fleet yourself, and each machine needs credentials to clone private repositories, either an SSH key or a GitHub App token; hosted macOS agents need Pro.

8. TeamCity (JetBrains)

TeamCity is JetBrains’ CI/CD server. Pipelines can be defined in a Kotlin DSL with IDE support, a natural fit for Java and Kotlin teams, and build chains model complex dependencies between builds.

TeamCity project overview showing build configurations, project health and a Flaky Tests tab

  • Best for: teams that want detailed build and test analytics, especially users of JetBrains IDEs such as IntelliJ IDEA.
  • Pricing: TeamCity On-Premises Professional is free for 3 build agents and 100 build configurations. TeamCity Cloud is $15 per committer a month, billed annually, with a 3-committer minimum (TeamCity pricing).
  • Pros: test history and flaky-test detection; remote runs test a change before you commit it; TeamCity 2026.2 made its newer, simpler Pipelines generally available.
  • Cons: build chains and the Kotlin DSL take time to learn; on-premises licensing gets complex at scale.

9. Harness CI

Harness CI is one module of the Harness platform, next to CD, security testing, infrastructure as code and feature flags. Test Intelligence runs only the tests a change affects.

Harness pricing page with Free, Essentials and Enterprise plans; the paid plans say Contact Sales

  • Best for: enterprises that want CI and CD under one governance model, with policy as code and audit trails.
  • Pricing: Free includes 2,000 Harness Cloud credits a month, about 2,000 minutes on a Linux 2-core machine; using Harness Cloud on Free requires a credit card. Essentials and Enterprise are sold through sales (Harness pricing).
  • Pros: strong guardrails, templates and role-based access across many teams.
  • Cons: more platform than a small team needs; no published prices for paid plans.

10. Google Cloud Build

Cloud Build is Google Cloud’s serverless CI/CD. A cloudbuild.yaml file defines the steps, and triggers start builds from GitHub, GitLab or Bitbucket. It builds container images and deploys to GKE, Cloud Run and other Google Cloud services.

Google Cloud Build product page stating that all customers get 2,500 build-minutes free per month

  • Best for: teams that build containers and deploy them on Google Cloud.
  • Pricing: 2,500 free build-minutes a month per billing account on the default e2-standard-2 machine, then $0.006 a minute (Cloud Build pricing). Private pools run builds inside your own VPC at separate rates.
  • Pros: no servers to run; tight links to Artifact Registry and Binary Authorization.
  • Cons: most useful only inside Google Cloud. Cloud Source Repositories, Google’s own Git hosting, has been closed to new customers since June 17, 2024, so your code will live elsewhere.

11. AWS CodePipeline and CodeBuild

On AWS, CodePipeline orchestrates the release stages and CodeBuild runs the builds and tests on managed machines. Both use IAM for permissions and can run builds inside your VPC.

  • Best for: teams that deploy almost everything on AWS.
  • Pricing: CodePipeline V2 charges $0.002 per action-execution minute after 100 free minutes a month; V1 pipelines are $1 per active pipeline a month. CodeBuild includes 100 free build minutes a month on its smallest machines, then from $0.005 a minute (CodePipeline pricing, CodeBuild pricing).
  • Pros: deep integration with S3, ECR, ECS, Lambda and CodeDeploy; costs follow usage.
  • Cons: setup is verbose, with IAM policies for each part. AWS CodeCommit, which stopped taking new customers in 2024, returned to general availability on November 24, 2025, so an all-AWS setup is an option again.

12. Semaphore

Semaphore is a hosted CI/CD service focused on fast pipelines, with parallel tests, promotions between pipelines and deployment targets. It connects to repositories on GitHub, GitLab and Bitbucket. Its Community Edition has been open source (Apache 2.0) since February 2025, so you can also run it yourself.

Semaphore home page showing deployment targets with status, who can deploy and the last deployment

  • Best for: teams that want fast hosted builds and simple usage-based pricing without per-seat fees.
  • Pricing: $15 of free credits a month (about 2,000 minutes on a 2-vCPU Linux machine). After that you pay per use with no seats: $0.0075 a minute for 2-vCPU Linux, billed per minute rounded up (Semaphore pricing). Pricing changed on March 16, 2026, dropping seat fees for self-hosted agents.
  • Pros: quick to set up; a self-hostable open-source edition.
  • Cons: a smaller ecosystem of integrations than the market leaders.

Free and open-source CI/CD tools

If cost is the constraint, the table above splits into three groups:

  • Open source, on your own servers: Jenkins, and Semaphore’s Community Edition (Apache 2.0). You pay for machines and upkeep instead of licenses.
  • Free hosted tiers: CircleCI has the largest allowance, up to about 6,000 minutes a month on a small Docker machine, then Google Cloud Build at 2,500 and GitHub Actions, Buildkite, Harness and Semaphore at about 2,000. GitHub Actions is free on standard runners for public repositories.
  • Cloud-native projects on Kubernetes: Tekton is an open-source framework for building CI/CD systems, and Argo CD does declarative GitOps delivery to Kubernetes. You run both on your own cluster, so they aren’t in the table.

How to evaluate and pick a CI/CD platform

Start with three questions:

  • Where does your code live? The platform built into your Git host (GitHub Actions, GitLab CI/CD, Bitbucket Pipelines) has the shortest setup and the best pull request integration. Move away only for a concrete reason: cost at your volume, a build that needs special hardware, or a compliance rule.
  • Where must builds run? Hosted CI on the vendor’s machines is the least work, so start there. If code or secrets can’t leave your network, as is common in regulated industries, use self-hosted runners on your Git host, Buildkite agents, Jenkins, TeamCity On-Premises or self-managed GitLab. Buildkite and self-hosted runners are hybrid setups, where the vendor hosts the control plane and your machines run the builds.
  • Where do you deploy? If nearly everything goes to one cloud, its native service (Cloud Build, CodePipeline, Azure Pipelines) saves IAM and networking work. If you deploy to several clouds, keep the pipeline on your Git host or another cloud-neutral platform so no single provider owns it.

Then estimate a month of real usage. Take last month’s builds, multiply by their average minutes and machine size, and price it on the two or three candidates; macOS minutes and large runners change the answer more than the headline price does. Run a one-week trial with your slowest pipeline before you commit.

Once it runs, measure it: how long a change takes from commit to production, how often deploys fail, how long a red build stays red. Our guide to DORA metrics and AI coding covers what to track. Coding agents make pipeline speed matter more: in one small team’s 866 agent-written pull requests, the median pull request went from opened to merged in 37 minutes, so a slow pipeline quickly becomes most of the wait. If agents write many of your pull requests, Claude Code analytics shows which sessions ended in a merged pull request and what they cost.

How we run CI for our own monorepo

We build Zest in one TypeScript monorepo, and its pull request checks run on GitHub Actions, on Blacksmith’s 2-vCPU runners rather than GitHub’s own. A script uses Turborepo’s change detection to work out which workspaces a pull request touched, so a change to one corner of the repository only lints, type-checks and builds that corner and whatever depends on it.

Between November 3 and 10, 127 pull request runs got through all of those checks. The median run took 2.5 minutes from being queued to finishing, and nine in ten finished within 4.8. On a sample of 35 recent runs, each one used about 7 runner-minutes across its jobs, which works out to about 4,000 minutes a month. At list prices that is about $24 on GitHub’s 2-core Linux rate and about $16 on Blacksmith’s $0.004 a minute. At our size the machine price hardly matters, and scoping the work matters a lot.

Blacksmith’s pricing page says its runners are twice as fast as GitHub’s, which we haven’t benchmarked. In our workflows a Blacksmith runner is just a different runs-on label. Depot sells the same kind of runner for GitHub Actions at $20 a month for 2,000 minutes, then $0.006 a minute. If your pipeline is already scoped and still slow, a faster runner is the first thing we’d try.

Frequently asked questions

How much do CI/CD platforms cost?

On the October 4 prices above, hosted Linux build minutes run from about $0.005 to $0.0075 on the usage-priced services (CodeBuild, GitHub Actions, Cloud Build, Semaphore). Per-user plans run from $3.65 a month on Bitbucket Standard to $29 on GitLab Premium and $30 on Buildkite Pro. GitLab Ultimate and Harness’s paid plans are sold through sales, with no published price. If you need a quote from those, or from Buildkite Enterprise or CircleCI Scale, bring your active-user count and last month’s build minutes so you can compare it with the published plans.

The billing units differ as well. GitHub Actions, Cloud Build and Semaphore bill per minute, GitLab and Bitbucket per user with minutes included, TeamCity Cloud per committer, CircleCI and Harness in credits, and Azure Pipelines per parallel job.

Is a free tier enough, or do I need an enterprise plan?

A free tier covers a small team until its build minutes run out. Move up when you need what free tiers leave out: GitLab keeps DAST, security dashboards and most compliance features in Ultimate, and Harness is built around policy as code and audit trails across many teams.

Harness or GitLab CI/CD?

GitLab CI/CD if you want source control, pipelines and security scanning in one product with published prices. Harness CI if an enterprise wants CI and CD under one governance model, with policy as code and audit trails, and doesn’t mind buying through sales.

Which CI/CD tools support Arm runners?

GitHub Actions has hosted Arm64 runners for Linux, Windows and macOS (labels such as ubuntu-24.04-arm), GitLab.com has hosted Linux Arm64 runners, and CircleCI offers Arm machines. Jenkins and Buildkite run on whatever hardware you give their agents, Arm included.

Which CI/CD platform suits an enterprise?

On GitHub, Enterprise Cloud adds SAML single sign-on and the largest included compute, $300 a month. GitLab Ultimate fits when one vendor should own security scanning and compliance. Harness suits a platform team enforcing policy as code across many teams, and Buildkite Enterprise adds SCIM, custom SAML and audit log exports while your builds stay on your machines. In a regulated industry, settle where builds must run first. We haven’t run all of these at enterprise scale, so treat this as a shortlist to trial, not a ranking.

What should a small team or startup use for CI/CD?

Use the one built into your Git host, on hosted runners, until a bill or a missing feature says otherwise. Without a dedicated DevOps engineer, running Jenkins means someone on the team also owns its upgrades and plugins. GitHub’s Free plan covers about 2,000 Linux minutes a month and CircleCI’s free tier up to about 6,000 for 5 active users. For scale, the pull request checks on our own monorepo come to about 4,000 runner-minutes a month.

What is the best CI/CD setup for a monorepo?

Pick the one that can skip what didn’t change. Buildkite has if_changed on steps and dynamic pipelines that generate steps at build time, and its docs cover Bazel. GitHub Actions filters workflows by paths, GitLab has rules:changes and Semaphore has change_in. Those decide which jobs run. A build tool such as Turborepo, Nx or Bazel decides what inside them is affected and caches the results, which is the setup we describe above.

Which CI/CD tools support GPU runners?

GitHub Actions has GPU-powered larger runners on Team and Enterprise Cloud, at $0.052 a minute on Linux and $0.102 on Windows for 4 cores. GitLab.com has GPU-enabled hosted runners on Linux for Premium and Ultimate, and CircleCI’s GPU executors run on Linux or Windows on its Scale plan. On Jenkins and Buildkite you attach your own GPU machines as agents.

Who offers hosted macOS build agents for CI?

GitHub Actions, Azure Pipelines and CircleCI all have hosted macOS machines, and GitLab.com’s are in beta for Premium and Ultimate. Buildkite’s hosted macOS agents run on dedicated Mac hardware and need its Pro plan, and you can also point its agent at Mac minis of your own. iOS builds are usually why a team needs these, and on GitHub a macOS minute costs far more than a Linux one, so price those builds first.

Is Jenkins obsolete, and what should we move to?

No. It still ships long-term support releases, the latest in October 2026. Move when its upkeep costs more than a hosted service would. The platform on your Git host is usually the shortest path: GitHub Actions Importer converts Jenkins pipelines, though scripted pipelines, secrets and unknown plugins have to be moved by hand, and GitLab publishes a Jenkins migration guide. Buildkite is the option if your builds must stay on your own machines.

Zest is AI coding analytics for engineering teams: it shows which coding agents, models and skills each engineer uses, what they cost, and which sessions end in merged pull requests. See how Zest works.