Claude Code vs GitHub Copilot (2026): Real Costs and Verdict

Both now have a desktop app, a terminal agent and a cloud agent. Here is what each costs in October 2026, what Claude inside Copilot really is, and a real task to test them on.

By Yotam DahanUpdated October 22, 202612 min readClaude Code

Claude Code vs GitHub Copilot: a real Claude Code session that built a bike shop status page, next to GitHub Copilot's Free, Pro, Pro+ and Max plan cards

Claude Code vs GitHub Copilot comes down to how you want to work and what you pay for. Claude Code is Anthropic's agent for handing over whole jobs, and a $20 Claude Pro plan runs it on Opus 5.5 by default; GitHub Copilot is cheaper to start (free, or $10 for Pro), lives inside GitHub and your code editor, and bills agent work in AI credits, with Claude's Opus models only from Pro+ at $39 a month. If you don't code, both now have a desktop app, but Copilot's asks for a GitHub account and Git, so Claude Code is the simpler first step.

Claude Code vs GitHub Copilot at a glance

The old split, autocomplete against agent, is gone: both now plan, edit files and open pull requests.

	Claude Code	GitHub Copilot
Desktop app	Claude app, Code tab	GitHub Copilot app (needs Git)
Free plan	No	Yes
Cheapest paid plan	Claude Pro, $20 a month	Copilot Pro, $10 a month
Claude Opus 5.5	Default on every paid plan	From Pro+ ($39), billed per token
Other companies' models	No	Yes: OpenAI, Google, xAI and more
Suggestions as you type	No	Yes, unlimited on paid plans

No independent benchmark compares them, because Copilot isn't in Artificial Analysis's Coding Agent Index or on Terminal-Bench 4.0; see our AI coding agents comparison for the tested ones.

What changed in GitHub Copilot in 2026

Older Copilot vs Claude Code comparisons still describe premium requests. 2026 changed that and more:

February 4, 2026Anthropic's Claude agent and OpenAI's Codex agent arrive on GitHub in public preview
April 20, 2026Claude's Opus models leave Copilot Pro
April 24, 2026Personal plans start sharing data for AI training unless you opt out
June 1, 2026GitHub AI Credits replace premium requests
June 17, 2026The GitHub Copilot desktop app becomes generally available
July 7, 2026The Copilot app reaches every plan, including Free

Dates come from GitHub's changelog and blog, checked October 22, 2026.

GitHub Copilot pricing and AI credits vs Claude plans

GitHub Copilot pricing now runs on AI credits, so sticker prices tell only half of the Claude Code vs Copilot pricing story:

Monthly price in USD, October 2026

Copilot Pro$10
Claude Pro$20
Copilot Pro+$39
Copilot Max$100
Claude Max 5x$100
Claude Max 20x$200

From GitHub's Copilot plans page and Claude's pricing page, October 22, 2026. Copilot Free costs $0; Claude's Free plan doesn't include Claude Code. Claude Pro is $17 a month billed yearly.

Here are the GitHub Copilot plans side by side:

Copilot plan	Price	AI credits a month	Claude models
Free	$0	Limited	Picked for you
Pro	$10	1,500 (worth $15)	Haiku and Sonnet, no Opus
Pro+	$39	7,000 (worth $70)	All, including Opus 5.5
Max	$100	20,000 (worth $200)	All, with priority access

Each paid allowance is a fixed base plus a "flex" part that GitHub says may change, reset monthly with no rollover (GitHub's Copilot plans). Claude Code comes with every paid Claude plan, compared in our Claude Code pricing guide.

GitHub Copilot pricing in AI credits: $15 of credits on Pro, $70 on Pro+ and $200 on Max each month, split into base credits and a flex allotment — One credit is $0.01 of model use, and the flex part can change. GitHub's Copilot plans page, October 2026.

What GitHub Copilot AI credits buy

Copilot turns the tokens each request uses into credits at the model's list price, which for Claude's models matches Anthropic's own API: Opus 5.5 costs $4 per million input tokens and $20 per million output tokens in both (GitHub's models and pricing, Claude's pricing page).

The math, with made-up numbers: a task re-reads 500,000 tokens it has seen (cached input), adds 100,000 new ones (billed as cache writes for Anthropic's models) and writes 40,000 back.

Claude Opus 5.5 (Pro+ and up): $0.10 + $0.50 + $0.80 = $1.40, or 140 credits, about 50 tasks on Pro+.
Claude Sonnet 5.5 (Pro and up): $0.05 + $0.25 + $0.40 = $0.70, or 70 credits, about 21 on Pro.

For a real number, Claude Code's /usage put our whole test session below at $0.56 in API terms, covered by a Claude plan's limits. The same tokens would be 56 Copilot credits, though Copilot's agent would use its own number of tokens.

Claude Code usage for one task: the /usage session block shows $0.56 at API prices and 190 lines added for the bike shop status page — What the whole session cost at API prices, Opus 5.5 plus a little Haiku 5.5, questions round included. Claude Code 2.1.303, October 22, 2026.

Is GitHub Copilot cheaper than Claude Code?

To start, yes: Copilot Free costs nothing and Pro costs half of Claude Pro. For Opus it's closer: Copilot bills every Opus token from a $39 plan, while $20 Claude Pro includes Opus 5.5 within its limits.

Claude in GitHub Copilot: what you get and what you don't

"Claude in Copilot", or GitHub Copilot Claude, can mean three different things, and only one of them is Claude Code:

Three ways Claude and Copilot meet

Claude models in Copilot

Pick Claude Sonnet or Opus in Copilot's chat, app or CLI. Billed in Copilot credits; Opus needs Pro+.

Anthropic's Claude agent on GitHub

A cloud agent built on Anthropic's Claude Agent SDK that turns an issue into a pull request. Public preview, paid from Copilot credits.

Claude Code itself

Anthropic's own agent in the terminal, the Claude app or VS Code. Runs on your Claude plan, never on Copilot credits.

GitHub's third-party agents docs and its plan cards list the GitHub Copilot Claude agent and the Codex agent from Pro up, while the plans page's comparison table marks them for Pro+ and Max, so check Copilot settings, Cloud agent, Partner agents first. The cards even label the agent "Claude Code", but the docs describe Anthropic's agent on the Claude Agent SDK, billed to Copilot credits.

A Claude Pro or Max login can't pay for the first two: Copilot bills them in its own credits, and your Claude plan only covers Claude Code itself.

Claude models in GitHub Copilot by plan: Opus and Fable models are missing on Free and Pro and included on Pro+ and Max — Opus left Copilot Pro in April 2026; Pro keeps Haiku and Sonnet. GitHub's Copilot plans page, October 2026.

A real task in Claude Code, and what to expect in Copilot

Riverside Cycle Repair is a made-up bike shop with 12 repair tickets in repairs.csv and prices in services.csv; the job was a page where customers look up their ticket. We ran it in Claude Code only; we did not run Copilot for this test, so its column comes from GitHub's docs. The prompt, with brackets for your own files:

The same job for both tools

In this folder, use [YOUR CSV FILES] to make [PAGE NAME].html: one search box where a customer types [WHAT THEY KNOW, FOR EXAMPLE A TICKET NUMBER] and sees [WHAT THEY NEED, FOR EXAMPLE STATUS, READY DATE AND PRICE]. It has to work offline on a phone. Ask me any questions first, then propose the steps and wait until I say go.

Claude Code 2.1.303 ran on Opus 5.5 at medium effort, in auto mode, where new terminal sessions start. My reply to its questions: "No questions from me, go ahead".

	Claude Code (our run)	Copilot CLI (from GitHub's docs)
Before editing	Asked 4 questions, proposed 5 steps, waited for "go" (23 seconds)	Plan mode asks questions and blocks edits until you accept the plan
Approval prompts	None: auto mode's classifier allowed each shell command	Asks the first time it needs each tool that edits or runs files
Time from "go"	81 seconds to its final message; status.html was on disk after 39	Not run
Files created	status.html, build_status.py (a rebuild script), status_template.html	Not run
Result	Worked offline; RC1046 showed "Waiting for parts", ready Wed 28 Oct, $40	Not run
What it used	$0.56 at API prices, 190 lines added	`/usage` shows AI credits per model

It tested all 12 tickets itself and said what it hadn't checked: "I haven't tried it on an actual phone." To test Copilot the same way, start in plan mode (Shift+Tab), approve one action at a time, and read /usage at the end.

The repair status page Claude Code built from two CSV files, showing ticket RC1046 with its status, ready date and price on a phone screen — The real output with made-up data, opened offline at phone width: RC1046, waiting for parts, ready Wed 28 Oct, $40. Built in 81 seconds, October 22, 2026.

Copilot cloud agent vs Claude Code in the cloud

Both can also work in the cloud on a GitHub repository. Copilot's version starts from an issue:

How Copilot's cloud agent handles one task

Issue
You write an issue or a prompt on GitHub
Assign
You assign it to Copilot, on Pro or higher
Cloud run
It works on its own branch, up to 59 minutes
Draft PR
It opens a draft pull request for you
You decide
Only a person can approve and merge

Each run covers one repository, one branch and one pull request (GitHub's docs). Claude Code's equivalents are cloud sessions (claude --cloud) and @claude in issues and pull requests, set up with /install-github-app. Without GitHub, skip this: both tools work on a plain folder.

Approvals, undo and privacy

	Claude Code	GitHub Copilot
Ask before each action	Manual mode	Asks per tool
Work on its own	Auto mode, the default	Autopilot, which you switch on
Undo a turn	`/rewind`, except files changed by shell commands	`/undo` or `/rewind`, for files Copilot changed
Trains on personal-plan data	Only if you allow it	Yes by default since April 24, 2026
Local session history	Kept 30 days on your computer	Synced to your GitHub account by default

Our test found a catch in undo. "Restore code and conversation" removed the two files Claude wrote directly, but status.html stayed, because Claude made it by running its rebuild script in the terminal, as the rewind menu warns.

Copilot's docs say its rewind restores only the files Copilot changed. Either way, keep a copy of the folder, or use Git, for anything you can't afford to lose.

Claude Code vs GitHub Copilot undo: the Claude Code rewind menu after building the bike shop status page, with options to restore code, conversation or both — Rewind undid the two files Claude wrote itself, but not status.html, which a shell command wrote: the menu warns about exactly that. Claude Code 2.1.303, October 22, 2026.

Before you point either tool at client work:

0 of 3 done

Training off in Claude's privacy settings and in your Copilot settings
Approvals on for the first week: Manual mode in Claude Code, no /allow-all in Copilot CLI
A copy of the folder before each big job, since neither undo covers everything

Copilot CLI vs Claude Code: the commands side by side

The two terminal agents now share most commands. In VS Code the closest match is Claude Code vs Copilot agent mode: both edit files across your project from a chat panel.

You want to	Claude Code	Copilot CLI
Start in a project folder	`claude`	`copilot`
Plan before changing anything	Shift+Tab or `/plan`	Shift+Tab or `/plan`
Let it work alone	Auto mode	Autopilot (`/autopilot`)
Undo the last turn	`/rewind` or Esc twice	`/undo` or Esc twice
See what it used	`/usage` (plan limits)	`/usage` (AI credits per model)
Write project instructions	`/init` writes CLAUDE.md	`/init` writes `.github/copilot-instructions.md`

Our guide to Claude Code plan mode explains when planning first pays off.

Use both on one project

Copilot CLI reads CLAUDE.md alongside its own instruction files, plus skills in .claude/skills. Claude Code doesn't read Copilot's .github/copilot-instructions.md, and its /import command doesn't cover Copilot.

So keep your rules in one CLAUDE.md (what goes in it), not copied into Copilot's file, or Copilot reads them twice. Add-ons work in both: see our MCP servers for VS Code and Copilot and GitHub Copilot skills. To run Claude Code in VS Code next to Copilot, follow our Claude Code tutorial.

Claude Code or Copilot: which to pick

Is Claude Code better than Copilot? For handing over whole jobs, usually; for suggestions while you type, Copilot is. Claude Code vs Copilot for beginners comes down to setup: Claude Code's desktop app needs no GitHub account or Git.

If this sounds like you	Pick	Why
You've never coded and want a site or tool built	Claude Code (desktop app)	No GitHub account or Git
You code every day in an editor	GitHub Copilot	Unlimited suggestions, from $10
Your team runs on GitHub pull requests	GitHub Copilot	Cloud agent and code review live there
You want Opus at a fixed price	Claude Code on Pro or Max	Copilot bills each Opus token
You want $0 first, or you're a student	Copilot Free or Student	Claude Code has no free plan
You want both	Copilot for typing, Claude Code for big jobs	One CLAUDE.md serves both

Quick check0 of 2 correct

What is the cheapest Copilot plan that lets you pick Claude Opus 5.5?

In our test, why did status.html survive Claude Code's rewind?

What to do next

Run the bike shop prompt, or your own version, in Claude Code this week on a copy of real files. If you have a GitHub account, run it in Copilot Free too, and keep the tool that needed fewer corrections.

If Claude Code wins, our Claude Code course takes you from setup to finished projects without assuming you code.

Claude Code and Copilot on one project: setup and cost kit

A CLAUDE.md both tools read, the privacy setting to check in each, a worksheet that turns Copilot credits into tasks, and five test tasks with a scoring sheet.

Enter your email and the full section opens right here. You also get a free lesson from the course.

Frequently asked questions

Is GitHub Copilot free?

Yes, Copilot Free costs $0 and includes 2,000 code completions a month, limited chat and agent use, the Copilot app and Copilot CLI, but not the cloud agent. Verified students get GitHub Copilot Student free, and verified teachers and maintainers of popular open source projects can get Copilot Pro free.

Does GitHub Copilot's Claude agent use my Claude subscription?

No. The Claude agent on GitHub is Anthropic's agent built on the Claude Agent SDK, in public preview, and it spends your Copilot credits, not your Claude plan. To use your Claude plan, run Claude Code itself, for example through Anthropic's VS Code extension.

What happened to Copilot premium requests?

GitHub replaced them with GitHub AI Credits on June 1, 2026: each interaction now costs the tokens it uses at the model's list price, and one credit is $0.01. People who kept an annual Pro or Pro+ plan stay on premium requests until it ends, then move to Copilot Free.

Does GitHub Copilot use my code to train AI?

On Copilot Free, Pro, Pro+ and Max, yes by default since April 24, 2026: GitHub may use your prompts, outputs and code snippets to train models. To stop it, open your Copilot settings and set 'Allow GitHub to use my data for AI model training' to Disabled; Copilot Business and Enterprise data isn't used for training.

Can I use Claude Code and GitHub Copilot in the same VS Code?

Yes. Anthropic's Claude Code extension and GitHub Copilot install side by side, and each signs in to its own account, so Claude Code uses your Claude plan and Copilot uses your Copilot credits. You could keep Copilot's completions while you type and give bigger, multi-file jobs to Claude Code.