What does Claude Code's GitHub Action cost? API usage plus Actions minutes
Claude Code's GitHub Action has two bills. Anthropic charges for the model, either through an API key at its per-token rates or, with an OAuth token, from your Claude subscription, in which case "runs use your Claude subscription instead of API billing". GitHub charges for the runner: Claude Code's docs say the action "runs on GitHub-hosted runners, which consume your GitHub Actions minutes", and in a private repository ubuntu-latest costs $0.006 a minute once your plan's included compute is used (both read on November 9, 2026).
Skipbox publishes this guide and sells a runner, which changes the minutes half only. The Anthropic and GitHub facts come from their own pages, read on November 9, 2026 and linked under Sources.
The two bills, side by side
| The model | The runner | |
|---|---|---|
| Who bills it | Anthropic (or Amazon Bedrock, Google Cloud or Microsoft Foundry, if you route the model there) | GitHub, for GitHub-hosted runners |
| What it counts | Input and output tokens of every turn | The job's running time, rounded up to whole minutes |
| How you pay | An API key (ANTHROPIC_API_KEY), billed per token, or an OAuth token (CLAUDE_CODE_OAUTH_TOKEN) that uses your Claude subscription | Your GitHub plan's included compute, then per minute |
| What makes it grow | Long prompts, a large codebase, many turns, a larger model | Long jobs, many runs, a larger runner |
| Where you see it | Anthropic's usage pages | GitHub's Billing & Licensing → Usage and the usage report |
The two grow together: a run with more turns takes longer, and the runner bills that time too.
What the model part costs
With an API key, Anthropic bills each run's tokens at the model's rates. Per million tokens, read on November 9, 2026:
| Model | Input | Output |
|---|---|---|
| Claude Opus 5.5 | $4 | $20 |
| Claude Sonnet 5.5 | $2 | $10 |
| Claude Sonnet 5 | $2 | $10 |
| Claude Haiku 5.5 (prompts up to 100,000 tokens) | $0.10 | $0.50 |
Cached input reads are billed at 5% of the input price on Opus 5.5 and Sonnet 5.5. Claude Code's docs say the action uses Claude Code's default model unless you pass --model in claude_args, and that "each interaction consumes tokens based on the length of prompts and responses, task complexity, and codebase size", so a run's cost depends on the request and the repository.
With an OAuth token from a Claude subscription, created with claude setup-token, runs use the subscription instead. Claude Code's docs list it on Pro, Max, Ultra, Team and Enterprise plans. For a secret shared across an organization's repositories, the docs recommend an API key, "since an OAuth token is tied to the subscription of the person who ran claude setup-token."
What the minutes part costs
The minutes part is GitHub arithmetic: the job's length × the runs a month × the runner's rate, less what your plan includes. GitHub rounds each job up to the next whole minute, and every run bills, including ones that end early.
The included compute, in dollars a month: $12 on GitHub Free, $18 on Pro and Team, $300 on Enterprise Cloud. Standard runners in public repositories are free. How many free minutes private repositories get has those dollars in minutes on each runner.
Examples in a private repository, on ubuntu-latest at $0.006 a minute unless the row names another runner:
| Workflow | Runs a month | Minutes per run | Minutes a month | Before included compute |
|---|---|---|---|---|
@claude on request, in comments | 100 | 8 | 800 | $4.80 |
| A review on every pull request event (opened, each push, reopened) | 400 | 6 | 2,400 | $14.40 |
| The same review on a Linux 4-core larger runner | 400 | 6 | 2,400 | $28.80 |
| A daily scheduled report | 30 | 5 | 150 | $0.90 |
A review workflow that triggers on synchronize runs on every push to a pull request. Claude Code's docs say it skips draft and closed pull requests and ones it already commented on, but the job still starts, and a job that runs 20 seconds bills a minute.
Copilot code review and Agentic Workflows use Actions minutes too
Claude's action isn't the only AI agent that runs on your Actions minutes now:
- Copilot code review, since June 1, 2026: "GitHub Actions minutes will be consumed from your existing plan entitlement", "with any usage beyond your included minutes billed at standard GitHub Actions rates." Public repositories are unchanged. GitHub's post says Copilot code review "also supports self-hosted runners and GitHub-hosted larger runners." From October 8, 2026, an organization can choose who is billed for the reviews themselves: the member's own Copilot entitlement (the default) or the organization, which needs AI Credits paid usage.
- GitHub Agentic Workflows, generally available on October 28, 2026: coding agents, Copilot or your own, run inside GitHub Actions workflows. "Because these are GitHub Actions, they reuse your existing runner groups and policy constraints." The default Copilot CLI engine needs a Copilot license.
So a repository with all three can run three agent jobs per pull request, each billing its own minutes.
Keep the minutes down
Claude Code's docs suggest specific @claude requests, a concise CLAUDE.md, --max-turns in claude_args, workflow timeouts, and GitHub's concurrency controls. In YAML, with the permissions the docs give for this workflow:
on:
issue_comment:
types: [created]
concurrency:
group: claude-${{ github.event.issue.number }}
cancel-in-progress: false
jobs:
claude:
if: contains(github.event.comment.body, '@claude')
runs-on: ubuntu-latest
timeout-minutes: 15
permissions:
contents: write
pull-requests: write
issues: write
id-token: write
actions: read
steps:
- uses: actions/checkout@v7
with:
fetch-depth: 1
- uses: anthropics/claude-code-action@v1
with:
anthropic_api_key: ${{ secrets.ANTHROPIC_API_KEY }}
claude_args: "--max-turns 10"
if:on the job keeps a runner from starting for comments that don't mention@claude. Without it, every comment starts a job.timeout-minutesstops a run that keeps going. GitHub's default is 360 minutes.--max-turnscaps the model's turns, which caps both bills.concurrencywith one group per issue or pull request runs one request at a time there: a new one waits while one runs, and by default GitHub keeps one waiting run per group and cancels an older waiting one. Setcancel-in-progress: trueto stop the running one instead.pathson apull_requesttrigger skips the review for changes it needn't read, such as docs-only pull requests.
The same cuts apply to every other workflow: GitHub Actions too expensive? finds the ones that cost the most and has eight cuts with the YAML.
The minutes half at a flat price
Every job picks its runner with runs-on, the job that runs Claude Code's action included. GitHub's docs: "Use jobs.<job_id>.runs-on to define the type of machine to run the job on", and "The destination machine can be either a GitHub-hosted runner, larger runner, or a self-hosted runner." GitHub bills nothing for self-hosted runner minutes, and they don't draw on your included compute.
Skipbox is a flat price for the runner minutes of Linux x64 jobs from private repositories: $39, $79 or $149 a month with unlimited minutes and no usage charges, each job in a fresh Ubuntu 24.04 VM in Helsinki. It covers the minutes half only; the Anthropic bill is the same whichever runner the job uses. Plan S is lower than ubuntu-latest on GitHub Team above about 9,500 job-minutes a month, and the cost calculator prices your own last 30 days of Linux x64 jobs by repository and workflow.
You can try it on your own jobs first: free for 7 days or 3,000 job-minutes, no card.
FAQ
Is Claude Code's GitHub Action free?
The action itself is open source (MIT), but each run has two costs: the model's tokens, billed by Anthropic through your API key or used from your Claude subscription when you authenticate with an OAuth token, and the GitHub Actions minutes of the runner the job runs on.
Can I use my Claude Pro or Max subscription with Claude Code GitHub Actions?
Yes. Claude Code's docs say a CLAUDE_CODE_OAUTH_TOKEN, generated with claude setup-token, authenticates with your Claude subscription on Pro, Max, Ultra, Team and Enterprise plans, and runs then use the subscription instead of API billing. For a secret shared across an organization's repositories, the docs recommend an API key, since an OAuth token is tied to the subscription of the person who created it.
Does Claude Code GitHub Actions use my GitHub Actions minutes?
Yes. Claude Code's docs say the action runs on GitHub-hosted runners, which consume your GitHub Actions minutes. In a private repository, ubuntu-latest costs $0.006 a minute once your plan's included compute is used; standard runners are free in public repositories.
Does the runner bill while Claude is thinking?
Yes. GitHub bills a job for the time it runs, rounded up to the next whole minute, and the job runs until Claude finishes. Time spent waiting on the model is runner time like any other.
Does Copilot code review use GitHub Actions minutes?
Yes, since June 1, 2026. Each Copilot code review in a private repository uses Actions minutes from your plan, and usage beyond the included minutes is billed at standard Actions rates. Public repositories are unchanged.
Sources
All read on 2026-11-09.
- Claude Code Docs, Claude Code GitHub Actions: the two kinds of cost, OAuth tokens and the plans that have them, the API key for shared secrets, the examples,
claude_args, the cost tips. - Anthropic, anthropics/claude-code-action: the MIT licence, the authentication methods.
- GitHub Docs, Choosing the runner for a job:
runs-onand the three kinds of runner. - Claude Docs, Pricing: per-token rates and cache reads.
- GitHub Docs, Actions runner pricing: per-minute rates, rounding up per job.
- GitHub Docs, GitHub Actions billing: included compute by plan, standard runners free in public repositories, self-hosted runners free.
- GitHub Docs, Workflow syntax:
timeout-minutesand its 360-minute default,paths. - GitHub Docs, Control the concurrency of workflows and jobs: one running and one pending run per group.
- GitHub Changelog, GitHub Copilot code review will start consuming GitHub Actions minutes on June 1, 2026, Copilot code review: new organization billing options and controls (October 8, 2026), and GitHub Agentic Workflows generally available (October 28, 2026).
Updated November 9, 2026