This article is about GitHub Copilot for coding, not Microsoft 365 Copilot in Word, Outlook and Teams; the office comparison is in Copilot vs Claude: What Works at the Office. Claude vs Copilot for coding, meaning Claude Code vs GitHub Copilot, comes down to this: Claude Code is Anthropic's agent that works in any folder, reads and edits files, runs commands and uses Claude models only; it needs a Claude plan from Pro (20 dollars a month, 17 with annual billing), but no GitHub. GitHub Copilot is GitHub's assistant: completions in the editor, chat, agent mode in the IDE, a cloud agent that prepares pull requests on GitHub, and Copilot CLI in the terminal. According to GitHub, Copilot also offers Claude models, Sonnet and Haiku on Pro, plus Opus and Fable from Pro+, and from Pro it includes a third-party agent from Anthropic in public preview. GitHub lists prices of 0 dollars (Free), 10 dollars (Pro), 39 dollars (Pro+) and 100 dollars (Max) a month, and 19 and 39 dollars per seat for companies (as of October 2026). If your team lives in GitHub and every change goes through a pull request, Copilot is the shorter route. If you want to work with Claude in folders and workflows, with or without a repository, Claude Code fits better.
I have used Claude Code every day since it launched, and my AI employees on Claude Code run on it. I have tested GitHub Copilot but never used it day to day. Everything about Copilot here therefore comes from github.com and the GitHub documentation, retrieved on October 26, 2026. The three-way comparison including Codex is in Claude Code, Codex or Copilot: Which One?; here I go deeper on Copilot and update it to October 2026.
Two approaches: an agent for your folder, an assistant for your repository
Anthropic's documentation describes Claude Code as an "agentic coding tool that reads your codebase, edits files, runs commands, and integrates with your development tools". You start claude in a terminal inside a folder, and the agent works there. On top come extensions for VS Code and JetBrains, a desktop app, sessions on the web and a GitHub Actions integration. You do not need a GitHub account to work in a folder.
GitHub's overview describes Copilot as an "AI assistant that helps you write, understand, and ship software". It suggests code as you type, answers questions about a codebase, reviews changes and works on tasks you assign it. Its focus is where GitHub itself is strong: issues, branches, pull requests and reviews. It does not work without a GitHub account.
For your decision: with Claude Code your folder is the workplace, with Copilot your repository on GitHub. If you do not build software but want an agent to work on quotes, reports or filing, Copilot is a detour; if you merge pull requests every day, Claude Code is one.
Claude Code vs GitHub Copilot: the comparison table
As of October 26, 2026, prices in US dollars per the vendors' own pages.
Claude models in GitHub Copilot
Yes, you can work with Claude in Copilot. Which models come with which plan is in GitHub's plan overview, as of October 26, 2026:
According to GitHub, Free and Student only get automatic model selection. Two things to know before you buy Copilot for Claude. First, GitHub bills models by the token, from your monthly credit. Per GitHub's price list, Opus 5.5 costs 4 dollars per million input tokens and 20 dollars per million output tokens, Sonnet 5.5 2 and 10 dollars, Fable 5.1 10 and 50 dollars, the same as with Anthropic directly. According to GitHub, Pro+ includes 7,000 AI credits a month in total, which is 70 dollars of credit at those prices. Once it is used up, you buy more. Second, Fable comes with a footnote: according to GitHub, Anthropic retains prompts and outputs by default for Fable 5 and Fable 5.1 to operate safety classifiers. Per GitHub, customers can request an exemption without retention until the end of 2026.
Which Claude model suits what, regardless of the tool, is in Claude Models: Opus, Sonnet, Haiku, Fable.
Agent mode, cloud agent and Copilot CLI per GitHub
GitHub Copilot now has three kinds of agents, and they are easy to mix up.
Agent mode in the IDE. Per GitHub, you can ask an agent to "find relevant code, edit files, run commands, and validate its work" while you watch and redirect it. According to the plan overview it is included in every plan, Free too. It is the closest to working with Claude Code, just in the editor rather than the terminal.
Cloud agent on GitHub. According to the GitHub docs, it researches the repository, plans changes and implements them in the background, in "an ephemeral cloud development environment", with its own branch and commits. Per the plan overview it is included from Pro and in Student, not in Free. GitHub has documented fixed guardrails for it: only people with write access to the repository can start the agent; it pushes only to its own branch; a human must review and merge its pull requests; by default, workflows only run after approval; internet access is limited by a firewall; and GitHub filters hidden characters out of issues as a defense against prompt injection. For teams that already work with branch protection, that fits the usual flow.
Copilot CLI in the terminal. Per the docs, you start it with copilot; it runs on Linux, macOS and Windows (PowerShell and WSL), has a plan mode, and -p for scripts. GitHub itself warns that with an automatic approval option such as --allow-all-tools, Copilot has the same access as you do. You turn on a local sandbox with /sandbox enable. According to the plan overview, the CLI is in every plan.
Claude Code covers the three cases differently. Locally in the terminal and the IDE it is the same tool. For work away from your machine there are cloud sessions on the web; according to Anthropic each one runs in its own virtual machine, GitHub credentials stay encrypted at Anthropic, and a proxy rejects branch deletions. For pull requests there is Code Review on Team and Enterprise, which I describe in Claude Code Review: Cost, Setup, Limits. And how to set the agent's permissions on your own machine is in Claude Code Security: Permissions and Sandbox.
Claude Code inside GitHub Copilot: the third-party agent
GitHub's plan page lists "Access to 3rd party agents (Claude Code and Codex)" for Pro. In the docs on third-party agents, the offering is called "Anthropic Claude coding agent", is marked as public preview, and "uses the Claude Agent SDK and can be powered by your existing Copilot subscription". You assign it an issue or give it a prompt; it works out the changes and opens a pull request.
According to the plan overview, third-party agents are available on Pro, Pro+ and Max, not on Free or Student; GitHub lists them for Business and Enterprise as well. One detail in the docs' model list stands out: for the Claude agent it names Auto, Sonnet 4.6, Sonnet 5, Opus 4.8, Opus 5 and Fable 5.1, but neither Opus 5.5 nor Sonnet 5.5. In practice, this is not the full Claude Code with your house rules, skills and hooks on your machine, but an agent inside GitHub's workflow built on Anthropic's Agent SDK. What the SDK is, I explain in Claude Agents: SDK, Managed Agents, Duet. If your company already pays for Copilot, it is a good way to try Claude as an agent before signing up for a plan of your own, with two caveats: the agent is in preview, and per the docs Opus 5.5 and Sonnet 5.5 are not yet available there.
Pricing and billing in October 2026
GitHub Copilot per the plan overview and github.com/features/copilot, as of October 26, 2026:
- Free: 0 dollars, 2,000 completions a month, an allowance of AI credits, automatic model selection only, Copilot CLI.
- Pro: 10 dollars a month, 1,000 base credits plus 500 "flex", 1,500 AI credits in total; unlimited completions, cloud agent, code review, third-party agents.
- Pro+: 39 dollars, 3,900 base credits plus 3,100 flex, 7,000 AI credits in total, plus Opus and Fable models and audit logs.
- Max: 100 dollars, 10,000 base credits plus 10,000 flex, 20,000 AI credits in total.
- Business: 19 dollars per seat, 1,900 credits per user in a shared pool; Enterprise: 39 dollars per seat, 3,900 credits.
According to GitHub, one AI credit equals 0.01 dollars, billed by model and token. GitHub explicitly calls the "flex" part variable, meant to adapt to model prices. Unused credits expire on the first of each month.
Claude Code is included from Pro per claude.com/pricing, as of October 26, 2026; prices are in the table above. Billing is not per token but through usage that renews every five hours, with weekly limits on top; only Fable is billed from usage credits on Pro and Team Standard. More in Claude Pricing and Plans Explained.
The difference in logic is bigger than the difference in price. With Copilot you know exactly how many dollars of credit you have, and every long agent session with Opus visibly draws it down. With Claude you have no dollar meter, but a time window. For long, continuous sessions like the ones my employees run, I prefer the fixed monthly price: a long session costs nothing extra as long as the usage lasts. Copilot, in turn, shows you more precisely what a session costs. If you only need an agent now and then and mostly want completions, Copilot Pro is cheaper.
Data use according to each vendor
Claude Code. According to Anthropic, Anthropic trains on data from Free, Pro and Max when that setting is on, Claude Code sessions included; the pricing page lists this as "Opt-out", so you switch it off yourself, much as with GitHub. Under commercial terms, meaning Team, Enterprise and the API, Anthropic does not train on code or prompts from Claude Code. Standard retention for commercial accounts is 30 days.
GitHub Copilot. For individual plans, meaning Free, Student, Pro, Pro+ and Max, GitHub writes on its plans page: "GitHub may use Copilot interaction data … to train and improve AI models." Per GitHub this means prompts, suggestions and code snippets from Copilot sessions. You can opt out at any time under github.com/settings/copilot/features. Per GitHub this applies "for Individual subscribers only"; for Business and Enterprise, GitHub refers to its Data Protection Agreement.
The same rule of thumb applies to both: for company code and customer data, use the business plan, not a personal account. What that means for Claude specifically is in Claude Privacy and GDPR: What Applies.
Which one fits
In my experience Claude Code fits when
- you want to work with Claude in folders, including without GitHub and without writing software,
- you run long, continuous sessions with large models and prefer usage limits to a dollar meter,
- you need house rules, skills and hooks for recurring workflows, the way I do for my workforce.
Based on GitHub's own descriptions, GitHub Copilot fits better when
- your team already works on GitHub and every change goes through a reviewed pull request,
- you want completions in the editor and only need the agent occasionally,
- you want to switch between models from different vendors, Claude included,
- you want to start free: Copilot has a Free plan, Claude Code does not.
You can also combine them: with GitHub Actions, according to Anthropic, Claude Code can automate code review and issue triage, so it works inside the GitHub flow without replacing Copilot. I treat it like any pair of tools: one writes, the other checks, never the same tool for both.
Frequently asked questions
Can GitHub Copilot use Claude models?
Yes. According to GitHub, Pro includes Haiku and Sonnet, Pro+ and Max add Opus and Fable, and Business and Enterprise also include Opus and Fable (as of October 2026). Usage is billed by token from your credit, at the same prices as with Anthropic.
Is Claude Code included in GitHub Copilot?
Partly. From Pro, GitHub offers an "Anthropic Claude coding agent" as a third-party agent in public preview, built on the Claude Agent SDK and running on your Copilot plan. That is not the same as Claude Code on your machine with your CLAUDE.md and your skills.
Which is cheaper, Claude Code or GitHub Copilot?
At entry level, Copilot: Free costs nothing, Pro 10 dollars. Claude Code starts with Pro at 20 dollars. The math changes once you work a lot with Opus: Copilot then charges for every million tokens, while with Claude you stay on your plan until you reach its limit. So there is no universal answer.
Claude vs Copilot: which is better for coding?
That depends on the workflow rather than the model, since both offer Claude models. If you work in GitHub with pull requests, Copilot gives you the built-in route. If you want an agent in your own folder with your own rules, Claude Code gives you that. I have not run a performance comparison of my own.
Is this the same Copilot as in Microsoft 365?
No. GitHub Copilot is for code, Microsoft 365 Copilot for Word, Excel, Outlook and Teams. The office comparison with Claude is in Copilot vs Claude: What Works at the Office.
Does Claude Code need a GitHub account?
No. Claude Code works in any folder. You only need GitHub for the GitHub integrations, such as GitHub Actions, Code Review or cloud sessions with a GitHub repository.
Where to go from here
If you want to try Claude Code, start with Claude Code: A Guide for Beginners. If you are weighing up editors, also read Claude Code vs Cursor: Which Fits You?. And if your company already has Copilot, test the Claude agent and the Claude models there first, before signing up for a second plan. How I run a whole workforce on Claude Code is what I show in my German-speaking community Claude Practitioners.
