GitHub Copilot Review: 4.7 Million Subscribers, One Costly Catch

Open your GitHub organization's billing dashboard in June 2026 and you might have needed a second look. One developer, posting in GitHub's own community forum, said a prompt that used to cost "barely a cent" the month before suddenly cost $4.51. That is not a typo. That is what happens when a tool used by 4.7 million paying subscribers switches from a flat subscription to metering by the token.

GitHub Copilot is Microsoft's AI pair programmer, built directly into nearly every code editor developers already use. It just went through the biggest pricing shake-up in its five-year history. This GitHub Copilot review covers what it actually does, what changed with billing on June 1, 2026, what every plan costs now, how it compares to Cursor and a few other names, and who should sign up. If you write code for a living, or you're about to start, this one is for you.

Quick Verdict

What works well:

• A genuine free tier: 2,000 completions a month, no credit card required, so you can try it before paying anything.

• Runs inside the editor you already use: VS Code, Visual Studio, JetBrains IDEs, Neovim, Xcode, and more, no need to learn a new interface the way Cursor asks you to.

• A real model picker. Send a task to Claude Opus, GPT-5, Gemini, or Grok, and switch depending on the job, all without leaving GitHub's ecosystem.

• The coding agent can pick up a GitHub issue, write the code, open a pull request, and even hand a task off to Claude Code or OpenAI Codex to execute while you keep working elsewhere.

• Built into GitHub itself, so code review, pull requests, and Copilot's suggestions all live in one workflow instead of three separate logins.

What to watch:

• Usage-based billing since June 1, 2026 means one long agent session can burn through the Pro plan's entire $15 in monthly credits, and overage charges stack on top.

• The Free plan's chat and agent access is described only as "limited." GitHub doesn't publish an exact number, so you won't know your ceiling until you hit it.

• Business and Enterprise pricing isn't posted publicly. You have to talk to GitHub sales to get a real number.

• There is no public referral or affiliate program for individual subscribers, only a separate "Partner Program" for companies building Copilot extensions. Nobody earns a commission for pointing you here, including us.

What Is GitHub Copilot?

GitHub Copilot is an AI coding assistant built by GitHub, which Microsoft owns. It launched in technical preview back in 2021 as a way to autocomplete code inside your editor, and it has since grown into something closer to a full AI teammate: it writes code, reviews pull requests, runs autonomously on tickets in the background, and now lets you choose which frontier model does the thinking.

By July 2025, Copilot had crossed 20 million all-time users, and Microsoft CEO Satya Nadella told investors on an earnings call that it had become "a larger business than all of GitHub was when Microsoft acquired it" for $7.5 billion back in 2018. Eighteen months later, Microsoft's January 2026 earnings call put paid subscribers at 4.7 million, up 75% year over year, with the pricier Pro+ tier growing even faster than that.

That scale matters here. This isn't a scrappy startup's side project, it's the default AI coding tool at roughly 90% of Fortune 100 companies. Which is exactly why the June 2026 billing overhaul made headlines: when a tool this deeply embedded changes how it charges, a lot of teams feel it at once, all at the same time.

Key Features

Copilot's feature set has expanded well past "finish my sentence." Here is what's actually in the product today, verified against GitHub's own docs and product pages:

• Code completion and Next Edit Suggestions. Predicts the next line, or the next edit, as you type. Unlimited on every paid plan, and it never touches your AI Credits.

• Copilot Chat. Ask questions about your codebase, get an explanation of a confusing function, or request a fix, right inside the editor or on GitHub.com.

• Agent mode. Give Copilot a broader instruction, like "add tests for this module," and it plans, edits multiple files, and runs commands on its own, checking in as it goes.

• Coding agent. A step further than agent mode: assign Copilot an actual GitHub issue and it works in the background, opens a pull request, and can even loop in Claude Code or OpenAI Codex as sub-agents when a task calls for it.

• Copilot code review. Automated review comments on pull requests, a useful first pass before a human reviewer looks at the diff.

• Model picker. Pick the model behind a given task from a genuinely wide bench. GitHub's own documentation lists options across OpenAI's GPT-5 family, Anthropic's Claude Opus and Sonnet lines, Google's Gemini, xAI's Grok, and Moonshot AI's Kimi models, rather than locking you to a single lab.

• Copilot CLI. Terminal-based help for developers who live in the command line more than an IDE.

• Copilot Spaces. A way to bundle docs, repos, and context into a shared reference that the whole team's agents can pull from.

Think of the model picker like a mechanic's toolbox instead of a single wrench. Pick Opus for the gnarly refactor, a cheaper mini model for boilerplate, and you're not paying frontier prices for busywork you didn't need a genius for.

How GitHub Copilot Actually Performs

Full transparency up front: this review wasn't written from inside a maxed-out Pro+ subscription run for months on end. What follows is grounded in GitHub's own documentation, its live pricing and model pages, and verified reporting on the June 2026 billing change, not an invented "we tested it for 30 days" story.

Here's what's verifiably true. Code completion and Next Edit Suggestions are the most mature part of the product. They've had five years to get good, and GitHub keeps them unlimited on every paid tier because that's the feature that made Copilot famous in the first place. Agent mode and the coding agent are newer, and that's exactly where the real complexity, and now the real cost, lives.

That's because agent sessions are what actually triggered the billing backlash. In GitHub's own developer community forum, engineers reported prompts jumping from fractions of a cent to several dollars overnight once usage-based Credits replaced the old flat "premium request" system on June 1, 2026. One AI Credit equals $0.01, and GitHub bills against real input, output, and cached tokens at each model's actual API rate. A long autonomous run on a heavyweight model like Opus can burn through a Pro plan's entire monthly allotment in a single afternoon.

In practice, that means the tool that used to feel like a flat-rate subscription now behaves more like a phone plan with data overage charges. Fine if you stick to light, everyday use. Easy to blow past if you let an agent run wild on a big task without checking in first. If your work is mostly quick completions and the occasional chat question, you probably won't notice much difference from before. If you lean hard on agent mode and the coding agent for multi-file, multi-hour tasks, budget for it and watch your usage dashboard. Don't assume the sticker price is the ceiling anymore, because right now, it isn't.

GitHub Copilot Pricing in 2026: What Changed, and What It Actually Costs

On June 1, 2026, GitHub replaced its old flat "premium request" allowance with usage-based AI Credits. The subscription prices themselves didn't change, but what you get for that money now depends on what you actually use. Here is every plan, verified directly on GitHub's live pricing and enterprise documentation as of writing:

One AI Credit equals one cent. Code completions and Next Edit Suggestions never consume credits, on any paid plan, so that part of the deal hasn't changed. Business and Enterprise numbers come from GitHub's enterprise billing documentation rather than the public plans page, so confirm current org pricing with GitHub sales before you commit a whole team to it.

Is it worth the price? For roughly the cost of a streaming subscription, Pro gives you unlimited completions plus $15 of credit toward chat and agent work, and that's a fair deal if you're not running agents constantly. Where it gets genuinely risky is Pro+ and Max, priced for heavy agent users but still capped, so one rough week of big autonomous tasks can mean an overage bill stacked on top of the $39 or $100 you already paid. Treat the credit allotment as a budget, not a subscription, and check usage before assigning Copilot anything big.

GitHub Copilot vs. the Competition

Copilot isn't the only AI coding tool that matured this year, and it isn't the cheapest one either. Here's how it lines up against the names you're probably also weighing:

We've already put Cursor through a full review of its own. If you're weighing "stick with the editor I already have" against "switch to an AI-first editor," that's the direct comparison worth reading next: our Cursor review. And if your real pain point is pull request review specifically, not writing code in the first place, Copilot's code review feature only partly covers that job. We go deeper on a tool built just for it in our CodeRabbit review.

Cursor forks VS Code and rebuilds the editing experience around AI from the ground up, which some developers find faster for big agentic tasks, at a higher $20 starting price and its own usage-based quirks. Devin Desktop, Cognition's June 2026 rebrand of Windsurf, leans hardest into full autonomy, useful if you want an agent that keeps working while you're away from the keyboard entirely. Amazon Q Developer and Tabnine solve narrower problems: AWS-native workflows and strict on-prem compliance, respectively, and are worth a look only if that specific need actually applies to you. For the wider field beyond just these five, we rounded up the AI coding platforms we've tried in one place.

Who Should Actually Use GitHub Copilot?

Best for:

• Developers and teams already living inside VS Code, Visual Studio, or a JetBrains IDE who want AI help without switching tools.

• Anyone who wants to test several frontier models, Claude, GPT, Gemini, Grok, without juggling separate subscriptions to each lab.

• Enterprises that need Copilot's audit logs, IP indemnity, and policy controls built into a platform they already use for source control.

Skip it if:

• You want one flat monthly price with zero risk of a surprise bill. The usage-based Credits system makes that hard to promise right now.

• You're chasing the newest, most experimental agent-first editing experience. Cursor and Devin Desktop currently push that frontier harder.

• Your only need is automated pull request review, not full coding assistance. A specialist tool will likely do that one job better, and cheaper.

The Verdict

GitHub Copilot earned its scale honestly. It's the AI coding tool with the deepest reach into the editors developers already use, backed by 4.7 million paying subscribers and adoption at nearly every major enterprise. That's not hype, that's Microsoft's own quarterly numbers talking.

But 2026 is also the year Copilot stopped being a predictable flat fee. The June shift to usage-based Credits is probably the right long-term call for GitHub, nobody can subsidize unlimited frontier-model agent usage at $10 a month forever, but it puts more responsibility on you to watch your own usage, especially once agent mode becomes a daily habit rather than an occasional convenience.

One honest note before you sign up: GitHub doesn't run a public affiliate or referral program for individual Copilot subscriptions, only a separate Partner Program for companies building Copilot extensions. So there's no special commission link to hand you here, just the real one: github.com/features/copilot. If you already code inside a supported editor and want AI help without leaving it, start on the free plan, watch how fast you burn through Pro's $15 in credits, and upgrade only once you actually know your usage, not before.