CircleCI Pricing Explained: How Credits Actually Work
This page answers the CircleCI pricing questions buyers actually ask — what a credit is worth, why the bill does not match the plan price, and where the credit model hides a five-fold cost jump. Figures are CircleCI list prices in USD.
The short answer: a credit is worth $0.0006 — 25,000 credits for $15. But “how many build minutes do I get” has no single answer, because consumption depends entirely on resource class. The same 30,000 included credits buy 6,000 minutes on Docker small or 150 minutes on macOS. And seats are also paid in credits from the same pool, which is why the bill rarely matches the headline plan price.
What are the plans?
| Plan | Price | Included credits | Included seats |
|---|---|---|---|
| Free | $0 | 30,000/month | Up to 5 active users |
| Performance | From $15/month | 30,000/month | 5 active users |
| Scale | Custom, billed annually | Custom | Custom |
| Server (self-hosted) | Custom | Unlimited build minutes | 30 seats included |
The “6,000 build minutes” headline on the Free plan is a best case. It is 30,000 credits divided by 5 credits per minute — which is only true on Docker small, a single vCPU. On Docker medium, the same allowance is 3,000 minutes. Quote the credits, not the minutes.
How much does each resource class actually cost?
| Docker (x86, Gen 1) | vCPU / RAM | Credits/min | $/min |
|---|---|---|---|
| Small | 1 / 2 GB | 5 | $0.0030 |
| Medium | 2 / 4 GB | 10 | $0.0060 |
| Medium+ | 3 / 6 GB | 15 | $0.0090 |
| Large | 4 / 8 GB | 20 | $0.0120 |
| X-large | 8 / 16 GB | 40 | $0.0240 |
| 2X-large | 16 / 32 GB | 80 | $0.0480 |
Docker Gen 2 runs a flat 20% premium over Gen 1 at every size for the same vCPU and RAM on paper — faster hardware, so cost per build may still fall, but cost per minute definitively rises. If a config migration flips your executors to Gen 2, the bill rises 20% with no visible signal in the config.
The Linux VM cliff — the biggest trap in CircleCI’s pricing
| Linux VM (Gen 1) | vCPU / RAM | Credits/min | $/min |
|---|---|---|---|
| Medium | 2 / 7.5 GB | 10 | $0.0060 |
| Large | 4 / 15 GB | 20 | $0.0120 |
| X-large | 8 / 32 GB | 100 | $0.0600 |
| 2X-large | 16 / 64 GB | 200 | $0.1200 |
Going from Linux VM large to X-large doubles the cores but quintuples the price — 20 to 100 credits per minute. Nothing in CircleCI’s interface flags this. A developer who bumps a machine executor one size up to fix a slow build has just taken a 5× cost increase, not a 2× one. This is the single most common cause of an unexplained bill jump, and it is worth putting a config policy around.
macOS, Windows and GPU
| Class | Spec | Credits/min | $/min |
|---|---|---|---|
| macOS m4pro.medium | 6 vCPU / 28 GB | 200 | $0.1200 |
| macOS m4pro.large | 12 vCPU / 56 GB | 400 | $0.2400 |
| Windows VM medium | 4 / 16 GB | 40 | $0.0240 |
| Windows VM large | 8 / 32 GB | 120 | $0.0720 |
| GPU large (Tesla V100) | 8 / 30 GB | 1,000 | $0.6000 |
For an iOS team the arithmetic is worth doing explicitly. A 15-minute build on m4pro.medium is 3,000 credits, or $1.80 per build. Forty builds a day across 21 working days is roughly $1,512 a month in compute alone, before seats. The Free plan’s entire monthly allowance buys ten such builds.
Charges that are not per-minute
- Docker Layer Caching: 200 credits per job — $0.12 per job, not per minute. A pipeline with 20 DLC-enabled jobs burns $2.40 before any compute.
- Storage and runner network egress: 420 credits/GB — about $0.25/GB.
- IP ranges: 450 credits/GB.
Why does the bill not match the plan price?
Because seats are paid in credits too, from the same balance as compute:
- Performance: 25,000 credits per active user per month — about $15
- Scale: 40,000 credits per active user per month — about $24
An “active user” is anyone who triggers compute on non-OSS projects — including commits that start builds, job re-runs, approving held jobs, and scheduled workflows and bot accounts. A release bot or a nightly pipeline consumes a seat.
Worked example. A 15-person team on Performance gets 5 free seats, so pays 10 × 25,000 = 250,000 credits ($150) per month in seat charges alone, before a single build minute. The 30,000 included credits cover roughly 1.2 seats’ worth. For teams with many committers and light CI, seats dominate the bill; for small teams with heavy macOS or GPU builds, compute does.
What forces the move to Scale?
| Free | Performance | Scale | |
|---|---|---|---|
| Concurrency | 30× | 80× | Unlimited |
| Self-hosted runner concurrent tasks | 5 | 20 | Unlimited |
| SSO / SAML | — | — | Yes |
| Audit log streaming | — | — | Yes |
| Config policies | — | — | Yes |
| IP ranges | — | Yes | Yes |
| Included storage | 2 GB | 2 GB | 200 GB |
| GPU classes | — | — | Yes |
SSO is Scale-only, and there is no self-serve path to Scale. An SSO requirement therefore forces an annual, sales-led contract regardless of your spend. That is usually the forcing function rather than concurrency.
One useful nuance: audit logs are available on all plans, retrievable for up to 12 months with a daily request limit. Only audit log streaming to a SIEM is Scale-gated. Compliance teams needing occasional evidence pulls do not need Scale; those needing continuous export do.
Do self-hosted runners cost credits?
Compute on self-hosted runners is free of credits. What still costs: network egress at 420 credits/GB for restoring caches and workspaces and downloading artifacts to your runners, storage at the same rate, and seats, which are unaffected by where compute runs.
This turns CircleCI into roughly a $15–24 per user per month control plane for a heavy CI shop. Two counterweights: the concurrency caps on runner tasks (5 on Free, 20 on Performance, unlimited on Scale) are the real gate, and on a cache-heavy monorepo the egress is not trivial — the more you self-host, the more cache traffic crosses the boundary.
Where CircleCI billing catches buyers out
Performance auto-refills instead of stopping
On Free, running out of credits means jobs fail until the next monthly allocation. On Performance, hitting zero triggers an automatic refill of roughly 25% of your subscription, minimum 25,000 credits — an automatic charge, not a hard stop. There is no true spending cap by default. If you are budgeting to a ceiling, that is the behaviour to change first.
Credits expire and are forfeited
Credits roll over month to month but expire one year after purchase, and unused credits are forfeited on cancellation. Prepaying for a volume discount carries real breakage risk if your usage is lumpy.
Bots consume seats
Scheduled workflows, release automation and service accounts all count as active users at 25,000 credits each per month. Organisations routinely under-forecast seats by not counting them.
How does this compare to GitHub Actions and GitLab?
| Workload | CircleCI | GitHub Actions | GitLab |
|---|---|---|---|
| Linux 2-core, per minute | $0.0060 | $0.0060 | $0.0100 |
| Linux 1-core, per minute | $0.0030 | Not offered | — |
| Windows, per minute | $0.0240 (4-core) | $0.0100 (2-core) | $0.0100 |
| macOS, per minute | $0.1200 (6-core) | $0.0620 (3–4 core) | Not on shared runners |
| Self-hosted compute | Free | Free | Free |
Three honest observations. On standard Linux, CircleCI and GitHub Actions are at price parity, and CircleCI is genuinely cheaper if your jobs fit in a single vCPU. Windows clearly favours GitHub. On macOS CircleCI looks roughly twice the price, but its smallest macOS class is six cores against GitHub’s three or four — per core they are comparable, and the real issue is that CircleCI has no small macOS SKU, so a light iOS job pays for six cores whether it uses them or not.
The structural difference matters more than the unit prices: GitHub bundles a minute allowance into a seat plan you are usually buying anyway for source hosting, whereas CircleCI’s seat fee is entirely incremental spend.
Frequently asked questions
How many build minutes do I get for $15?
5,000 Docker-medium minutes, or 125 macOS minutes. There is no single answer, which is precisely what the credit abstraction obscures. Price your actual pipeline mix.
What happens when we run out of credits mid-sprint?
On Free, jobs fail until the next allocation. On Performance, credits auto-refill and you are charged. Without auto-refill configured, builds are blocked until the next allocation date with the error “This job has been blocked because no credits are available on your plan.”
Is CircleCI Server still available?
Yes, and actively developed — version 4.10 shipped in May 2026, with a 5.0 release signalled. It includes 30 user seats and unlimited build minutes with no credit metering on compute. Pricing and the licensing model are not published.
Why did our bill jump when we resized a machine?
Almost always the Linux VM large-to-X-large step: 20 to 100 credits per minute, a 5× increase for 2× the cores.
Do inactive users cost us anything?
No — a developer who triggered no compute that month is not billed. But bots, scheduled workflows and service accounts do count, which usually more than offsets it.
Can we cap our spend?
Not by default on Performance, because auto-refill is the standard behaviour. Controlling spend means disabling auto-refill and accepting blocked builds, or moving compute to self-hosted runners and paying mainly for seats. Which is right depends on whether a stopped pipeline or an unexpected invoice is the worse outcome for you.
Pricing verified 19 September 2026 against CircleCI’s published pricing page, price list and documentation. Dollar figures are derived from CircleCI’s published credit rates at $0.0006 per credit; CircleCI publishes seat costs in credits rather than dollars. Scale and Server pricing are not published.
Work out what your pipeline actually costs
We model CI spend against real pipeline data — resource class mix, seat counts including bots, and whether moving compute to self-hosted runners pays for itself. It is usually a different answer than the plan page suggests.