GitHub Copilot vs Claude Code
By Mari Luukkainen · 6 min read · Last updated
On this page
GitHub Copilot is an assistant built into your editor and GitHub, with completions as you type, chat, an agent mode, a command line tool and a coding agent that turns issues into pull requests. Claude Code is Anthropic's agent that takes a whole task. Pick Copilot for help inside the editor you already use, Claude Code for delegated multi-file work.
The two overlap more than they did, so this page compares habits, instruction files, plans and the pull request workflow, with prices checked on 1 November 2026. I have not written about using Copilot, so what follows comes from GitHub's and Anthropic's documentation. See also the tool overview.
Ask AI about this pageChatGPTClaudePerplexityGoogle AI ModeGrok
What does each tool do day to day?
Copilot lives where you already work. In VS Code and the JetBrains editors it suggests the next lines while you type, answers questions in a chat panel and makes edits across files in agent mode. Its command line tool, Copilot CLI, works in the terminal like Claude Code does. On GitHub it can be given an issue, work on it and open a pull request for you to review.
Claude Code reads the repository, edits files, runs commands and reports. It runs in the terminal, VS Code and JetBrains, a desktop app and on the web. It has no typing suggestions.
| Moment | Copilot | Claude Code |
|---|---|---|
| You are typing a function | Suggests the rest of the line or block | Not involved |
| You have a question about a file | Chat panel, with the file as context | Ask in the terminal, it reads what it needs |
| You want a feature across six files | Agent mode in the editor, or Copilot CLI | Describe it once, review the diff |
| You want a task done while you do something else | Assign an issue on GitHub, review the pull request | A cloud session on the web pushes a branch and can create a pull request |
| Which models | A model picker that includes Claude models, and an auto selection on the free plan | Claude models |
| Project instructions | .github/copilot-instructions.md, and it reads AGENTS.md, CLAUDE.md and GEMINI.md | CLAUDE.md, or AGENTS.md when there is no CLAUDE.md |
Prices (checked 1 November 2026)
| Plan | Price per month | What GitHub lists |
|---|---|---|
| Copilot Free | $0 | 2,000 completions a month, limited chat and agent usage, automatic model selection only |
| Copilot Pro | $10 | Unlimited completions, 1,500 GitHub AI Credits (1,000 base and 500 flex) |
| Copilot Pro+ | $39 | Unlimited completions, 7,000 GitHub AI Credits (3,900 base and 3,100 flex) |
| Claude Pro, for comparison | $20 ($17 billed annually) | Includes Claude Code. Max is $100 and Ultra $200 |
Usage is counted in GitHub AI Credits, where one credit is $0.01. This replaced the older premium requests. Sources: GitHub plans, claude.com/pricing. Copilot Free is the cheapest way to try an AI assistant in an editor. Claude Code has no free plan.
Which suits a vibe coder with no coding background?
Neither is the gentlest start. For someone who has never written code, a browser builder is easier. Between these two, Copilot's chat is easier to try if you already use VS Code, and Claude Code needs comfort with a terminal. Once you can read a diff, the agent style gives more output per prompt.
My own start was in an editor. In March 2025 I was a VC, saw vibe coded products in our dealflow, and asked ChatGPT what this was. It pointed me at Cursor. I built my first Next.js app in Cursor on a two hour flight from Helsinki to Berlin, having never used GitHub (post). An editor with a chat panel, which is what Copilot gives you in VS Code, matches that path.
The pull request workflow
The workflow that sets Copilot apart is on GitHub: write an issue, assign it, review a pull request. It works when issues are written clearly. This format also works as a prompt for Claude Code:
Title: Show an empty state on the orders page
Problem: when a customer has no orders, the page shows a blank area.
Done when:
- the orders page shows the text "No orders yet" and a link to /shop
- the empty state has a unit test
- no other page changes
Files to look at first: app/orders/page.tsx, components/OrderList.tsxThe "Done when" list turns your review into a checklist. Naming the files to look at first saves the agent from searching. Claude Code can do the same work: a cloud session pushes a branch and you create the pull request from claude.ai/code, or you mention @claude in an issue with the GitHub Actions integration.
What should repository instructions say?
Copilot reads .github/copilot-instructions.md. GitHub's documentation also says you can create AGENTS.md files anywhere in the repository, and the VS Code setting chat.useAgentsMdFile is on by default (GitHub docs). The simplest setup for both tools is AGENTS.md as the source, and a CLAUDE.md containing the single line @AGENTS.md. The content follows the rules for CLAUDE.md: short, concrete, from real mistakes.
This is a Next.js project with TypeScript.
Run npm test and npm run build after changes.
Use the existing components in components/ before creating new ones.
Do not add dependencies without asking.
Write tests next to the file they cover, named *.test.ts.
Never include secrets or .env contents in code or comments.A line in an instructions file does not block anything. To keep an agent out of a file, use a permission rule, as in Claude Code best practices.
How do you review an agent's pull request?
- Read the description. Does it match the issue, and does it say what was not done?
- Check the file list. Files you did not expect are the first thing to question.
- Open the tests. Were tests added or edited, and were existing tests changed to pass?
- Check the CI result, then run the change yourself on a branch.
- Look for new dependencies in package.json and the lock file.
- Search the diff for keys, tokens and hard-coded addresses.
The same list works for a diff from any agent.
Which should you choose?
| If | Choose |
|---|---|
| You work in VS Code or JetBrains and want help while typing | Copilot |
| Your team lives in GitHub issues and pull requests | Copilot |
| You want to try an AI assistant without paying | Copilot Free |
| You want to give a task and review a full diff | Claude Code |
| You are comfortable in a terminal | Either: Copilot CLI or Claude Code |
| You want both | Use both, and commit between them |
Compare further in Cursor vs Claude Code and Codex vs Claude Code. The alternatives list covers the rest.
Frequently asked questions
Can GitHub Copilot use Claude models?
Yes. GitHub lists Claude models as generally available in Copilot, including Haiku, Sonnet, Opus and Fable versions, and Fable needs an administrator to enable it. I could not confirm which plan includes which model, so check the supported models page.
Is GitHub Copilot open source?
The VS Code Copilot Chat extension was open sourced under the MIT licence, announced on 30 June 2025. The Copilot service itself is not open source.
How do I install the Copilot command line tool?
Run npm install -g @github/copilot, brew install --cask copilot-cli, or winget install GitHub.Copilot. It needs an active GitHub Copilot subscription.
What are GitHub AI Credits?
The unit Copilot usage is counted in since the plans changed. One credit is $0.01. Copilot Pro includes 1,500 a month and Pro+ includes 7,000.
Next guideOpenCode vs Claude Code →