The automated jobs are multiplying. Tests, builds, recurring scripts, maybe a batch of data that needs sorting every night. Then comes the dreaded message: the free CircleCI allowance is nearly gone.
Time to pay? Maybe. But looking for another place to run the work produces an unexpected problem: there are almost too many volunteers.
It's like posting one vacancy for a computer and watching an entire waiting room fill up. Nobody has agreed to work unlimited hours, but the queue is impressive.
Meet the applicants
A runner is simply a computer that executes a job for you. Instead of leaving your laptop on all night, you send the task to a hosted service.
Google Cloud Build offers 2,500 free build-minutes a month on an eligible default-pool machine. It is a natural candidate for software builds and tests. Google Cloud Run Jobs has a different allowance: 240,000 free vCPU-seconds per month, roughly 4,000 minutes at one vCPU, with a separate memory allowance and regional pricing details. It is designed for batch jobs rather than conventional CI alone.
Azure Pipelines offers eligible private projects a Microsoft-hosted free tier of 1,800 minutes per month, once enabled, with a 60-minute limit per job. Buildkite lists 2,000 free Linux vCPU-minutes monthly; on a two-vCPU runner, that is about 1,000 minutes of elapsed run time. GitLab Free provides 400 compute minutes. Bitbucket Free provides 50 build minutes. Even the smallest applicant has a role.
GitHub Actions makes standard hosted runners free in public repositories, while private repositories on GitHub Free receive an allowance of 2,000 minutes monthly. But its terms tie hosted-runner usage to building, testing, deploying or publishing the related software project. It is not a general-purpose unlimited free computing farm.
Cloudflare Workers offers up to 100,000 requests a day on its free plan, but only 10 milliseconds of CPU time per invocation. Think doorbell, dispatcher or alarm clock, not a heavyweight overnight employee. Oracle Cloud's Always Free virtual machines offer another route for people willing to maintain their own server.
The interview has gone well. Unfortunately, the recruiter now needs a spreadsheet.
Minutes are not all the same minutes
This is the detail that ruins the tempting arithmetic. A build-minute, a compute minute, a vCPU-minute and a vCPU-second are not interchangeable.
A two-vCPU job can consume twice as many vCPU-minutes as minutes on the clock. Other services charge separately for memory, storage, network traffic, logging or scheduling. Google Cloud services can require a billing account even when the usage stays within a free tier. Cross the limit and the bill may not stay at zero.
Adding every headline allowance into one heroic total is like adding kilometers, kilograms and cups because they all have numbers. It looks excellent until someone asks what the total means.
The useful trick is dividing jobs, not multiplying accounts
Keep one reliable home for the code and results. Let different providers execute different authorized jobs. Builds may suit Cloud Build; longer scheduled batches may suit Cloud Run Jobs; a lightweight trigger may suit Cloudflare Workers; remaining CircleCI capacity can still do real work.
Then check three things: how much capacity is actually left, whether a job is running twice, and whether failures leave useful records. Three providers performing the same task three times is not clever scaling. It's ordering the same dinner three times and congratulating yourself on the menu variety.
So yes, CircleCI reaching its limit does not mean a project has reached its limit. More free options exist than many developers expect. Read the current terms before deploying anything.
The punchline: the computers may work for free, but their shift manager does not.
Official references, checked 12 October 2026: CircleCI, Cloud Build, Cloud Run, Azure Pipelines, Buildkite, GitLab, GitHub terms, Cloudflare limits.
