GitHub Copilot Pricing: What You Actually Pay

What does GitHub Copilot cost?

GitHub Copilot Pro is $10/month, which remains the lowest entry price of any serious AI coding tool. Pro+ sits above it, and there are Business and Enterprise tiers for organisations. The important change is not the price — it's what the price now buys. Since 1 June 2026, Copilot's allowance is denominated in GitHub AI Credits rather than a monthly count of "premium requests". Pro includes $15 of credits per month, Pro+ includes $70, where 1 credit = $0.01. If you see a comparison quoting "300 premium requests per month", it's describing billing GitHub has retired — and that includes several round-ups still published as current. The figures here were checked against GitHub's own pricing and documentation pages rather than secondary sources, for exactly that reason.

What's actually unlimited?

Code completions and next-edit suggestions don't draw on credits at all. GitHub is explicit that they remain unlimited on every paid plan. This matters more than it sounds, because completions are what most people use Copilot for most of the time. If your workflow is "type, accept the ghost text, keep going", you are using the unmetered part of the product and your $10 buys genuinely unlimited use of it. What consumes credits is the agentic surface: chat with premium models, agent mode, code review, and the heavier multi-file operations. That's the part where a bill can move, and the part worth understanding before you scale usage.

How do credits actually get spent?

A credit is $0.01, and different operations draw different amounts depending on which model handles them. The practical consequence is that the same $15 stretches very differently depending on your habits. Three patterns, roughly:

Completion-heavy work. You may never touch your credit balance. The metered surface simply isn't where you spend your time.
Chat-heavy work on premium models. Credits deplete at a pace that surprises people who came from the request-counting era, where a "request" was a flat unit regardless of how much computation it triggered.
Agent mode on large repositories. This is where the meter runs fastest, because an agent re-reads its accumulated context on every step — the same compounding that drives cost on every agentic tool, not a Copilot quirk. How the context window drives your token bill covers the mechanism.

The shift from requests to credits is, in fairness, more honest: you pay for computation performed rather than for round-trips initiated. It's also less predictable, which is the trade-off.

Is Copilot cheaper than Cursor or Claude Code?

At the entry price, yes — $10 against $20 for both. Whether it's cheaper for you depends on which surface you live in. The comparison people get wrong is treating these as three prices for one product. They aren't:

Copilot is strongest as an in-editor assistant across many IDEs, with unlimited completions and a metered agentic layer on top.
Cursor is an AI-native editor — you switch editors to get it, and the agent is central rather than an add-on. Cursor pricing covers its meter.
Claude Code is a terminal agent with the deepest reasoning of the three, bundled into a Claude subscription rather than sold separately.

If you want AI in the editor you already use and mostly want completions, Copilot at $10 is hard to beat and the unlimited completions are real. If you want an autonomous agent doing multi-file work, the $10 buys you the least of the three, because that's the metered surface.

Where does the money actually go?

On the agentic surface, cost is driven by tokens, and tokens are driven by how much context each step carries — not by how many messages you send. This is the part that survives any pricing model change, which is why it's worth understanding rather than memorising a plan table. An agent that reads six files to answer a question about one costs more than an agent that reads one, on every tool, under every billing scheme. Credits, requests and per-token API pricing are three ways of denominating the same underlying consumption. Four levers move it, and none is Copilot-specific:

Narrow what the agent reads. Point it at a symbol, not a directory.
Filter noisy command output before it lands in the conversation.
Keep sessions bounded — one task per session, so early reads don't get re-billed all afternoon.
Prune connected tools. Every MCP server advertises its schemas every turn, invoked or not.

Tokenade applies these automatically across Copilot, Claude Code, Cursor, Codex and Windsurf, with measured savings on a dashboard rather than a claim. The manual route works too — it just requires doing it every day.

Which plan actually fits you?

Start from which surface you use, not from the price ladder. The plans differ almost entirely in metered allowance, and the metered surface is the one many people barely touch. Free tier. Enough to decide whether the completions suit your language and codebase, which is the thing worth evaluating. Not enough to judge what sustained agentic use costs. Pro at $10. The right default for the large majority. If your day is completions plus occasional chat, you are buying unlimited use of the part you actually use, and $15 of credits is generous headroom for the rest. This is the plan most people should start on and stay on. Pro+ at the higher tier. Justified when you consistently exhaust credits on the metered surface — heavy agent mode, heavy premium-model chat, code review at volume. The honest test is whether you hit the ceiling more than once in a month. Once is an anomaly; every month is a signal. Business and Enterprise. These change governance more than they change economics: policy controls, audit, seat management, and the data-handling terms an organisation needs. Evaluate them on compliance requirements, not on credit arithmetic. One pattern worth naming, because it's common and expensive: people upgrade after a month where agent mode ran wild, when the actual fix was a bounded session and a narrower read. The upgrade works — it raises the ceiling — but it also removes the signal that something was inefficient. If you're about to move up a tier, spend ten minutes finding out what consumed the credits first. If the answer is "one afternoon of agent mode on a big repository", the upgrade is treating a symptom you could remove. The corollary is that the cheapest plan is not always the cheapest outcome. Rationing a $10 plan while fighting the ceiling costs time, and time is the expensive resource. Pick the plan that matches how you actually work, then reduce consumption so that plan keeps fitting.

What goes wrong (anti-patterns)

Budgeting from a retired figure. "300 premium requests" describes billing GitHub replaced on 1 June 2026. Any guide still quoting it hasn't been updated, which tells you something about the rest of its numbers. Assuming everything is metered. Completions aren't, on any paid plan. People sometimes ration the free part while burning the metered one. Upgrading to escape a context problem. More credits let you waste more before stopping. If agent mode is draining your balance, look at what it's reading first. Comparing on entry price alone. $10 versus $20 is meaningless without asking which surface you'll actually use — the cheapest plan for completions is not the cheapest plan for agents.

Frequently asked questions

Is GitHub Copilot free for anyone?

GitHub has historically offered free access to verified students, teachers and maintainers of popular open-source projects, and there is a limited free tier. Check GitHub's current eligibility page rather than a third-party summary — this is exactly the kind of detail that changes without comparison articles noticing.

What happens when I run out of credits?

You keep unlimited completions, since those never drew on credits. The metered features stop until the next cycle unless your plan or organisation allows additional spend. This is softer than a hard lockout, and it's another reason completion-heavy users rarely notice the meter.

Do credits roll over?

Treat monthly allowances as use-it-or-lose-it unless GitHub's current terms say otherwise. Don't plan around rollover you haven't verified on the vendor's page.

Is Copilot Pro+ worth it over Pro?

Only if you're consistently hitting the credit ceiling on the metered surface. Pro+ multiplies the allowance considerably, but if you're exhausting $15 of credits because agent sessions carry bloated context, the upgrade buys you more of the same waste. Fix the consumption, then size the plan.

Does Copilot cost extra on top of the subscription?

Not for completions. For metered features, your plan's included credits are the budget; going beyond them requires either a higher plan or, in organisational setups, permitted overage. There's no separate token bill the way there is with direct API use.

See also:

Cursor pricing: what you actually pay — the other $20 option, metered differently.
Claude Code pricing — subscription tiers against API rates.
Best AI coding tools — where Copilot sits against the field.
LLM cost: what drives it and how to cut it — why the tariff is the least interesting variable.