GitHub Copilot

Where GitHub Copilot sits

Copilot is the default AI coding tool for a large share of professional developers, and the reason is distribution rather than any single feature: it is inside VS Code, Visual Studio, the JetBrains IDEs, Xcode, GitHub.com, GitHub Mobile, the Windows Terminal and a standalone desktop app and CLI — all plans now include Copilot CLI and the Copilot app (docs.github.com/en/copilot/get-started/plans). If your code lives on GitHub, Copilot is already wired into the surface where you review pull requests, triage issues and write code. That network effect is the product.

The billing model changed this year and it is the first thing to understand. On 2026-06-01, all Copilot plans moved to usage-based billing: premium request units were replaced by GitHub AI Credits, consumed by token usage — input, output and cached tokens — at each model's listed API rates (github.blog, 2026-04-27). Base prices did not change: Pro is $10/month, Pro+ $39/month, Business $19 per seat, Enterprise $39 per seat. Every plan now ships a monthly credit allowance, and paid plans can buy more.

What it actually does

There are two layers of assistance. The first is the in-editor loop: code completions and Next Edit Suggestions, plus Copilot Chat with inline chat, slash commands and chat skills in VS Code and Visual Studio. Critically, completions and next-edit suggestions remain included on every plan and do not consume AI Credits — so the passive, constant layer of help is unmetered. Free accounts get 2,000 completions per month and auto model selection only.

The second layer is agentic, and it is where the credits go. Agents on GitHub take an assigned task — an issue, a bug report — work in the repository and open a pull request for review. The coding agent is available on paid plans and limited on Free; Copilot Student excludes third-party agents. On GitHub itself, chat lives on pull requests and issues; in the terminal, Copilot CLI handles shell and git workflows. Model choice is wide on paid plans — the docs list Claude Haiku 4.5 and 5.5, Claude Opus 4.8 (with a fast-mode preview), Claude Opus 5, and GPT-family models among others — with priority access to premium models on Max and Enterprise. Enterprise admins control which models are enabled, and the docs are unusually explicit about data handling: for Claude Fable 5 and 5.1, Anthropic retains data by default for safety classifiers, with zero-data-retention access available to approved enterprises through the end of 2026 under a time-bound exemption.

My testing notes

The most useful fact from my testing is how the credit meter behaves. In-editor completions churned all day without moving my credit balance, while a single agentic task — "fix this failing test and update the caller" — consumed a visible slice of credits because it burned tokens across many file reads and edits. That split makes Copilot's economics easy to live with if you use the agent sparingly and expensive if you delegate whole features to it. The allotments at time of writing: Pro includes 1,000 base plus 500 flex credits (1,500 total), Pro+ 3,900 base plus 3,100 flex (7,000), Max 10,000 base plus 10,000 flex (20,000), Business 1,900 per user and Enterprise 3,900 per user.

Second, the coding agent's GitHub-native loop is its strongest workflow. In my testing, assigning a well-scoped issue produced a branch and a pull request with the change and a written summary, ready for review in the same interface I would have used anyway. It is not the fastest agent I have used, and it occasionally chose a more conservative change than I wanted, but the review-and-merge cycle on github.com is where I spend my time, so friction there matters more to me than raw generation speed.

Third, Free is a real tier rather than a demo. Two thousand completions a month with auto model selection covers students and casual use, and verified students get Copilot Student free; verified teachers and popular open-source maintainers can get Pro free. For a beginner, that is enough assistance to learn with, at zero cost.

For organisations, the Business and Enterprise tiers are mostly about governance. Centralised seat management, policy control over which models and features members can use, and — on Enterprise — a larger credit pool (3,900 per user versus 1,900 on Business) plus enterprise-grade controls, with GitHub Enterprise Cloud as the required substrate. In my testing of the admin surface, policy enforcement is straightforward: an administrator can enable or disable individual models across the org, which is exactly the control point you want when a model's data-retention terms differ from the others. Copilot's activity also feeds naturally into review culture, since agent-authored pull requests go through the same checks and approvals as human ones.

A practical budgeting note from living with the meter: model choice dominates cost. Running premium models on every agent turn empties a Pro allowance quickly, while routing routine edits through auto model selection and reserving premium models for hard problems stretches the same 1,500 credits much further. The blog post's framing — pricing aligned to actual usage instead of gating heavy users — is accurate as far as it goes, but it also means your monthly AI bill is now variable in a way it was not under flat premium-request counting. Watch the billing overview page for the preview-cost tooling GitHub added before the transition.

Where it falls short

The June 2026 billing change removed one of my favourite safety nets: when you exhausted premium requests, Copilot used to fall back to a cheaper model and keep working. Fallback experiences are gone; when your credits run out, you buy more or wait. Heavy agent users on Pro should expect top-ups, and annual Pro/Pro+ subscribers are in a transitional state — they stay on premium-request pricing until expiry (with increased model multipliers), then drop to Copilot Free unless they re-upgrade, which caught people I know by surprise.

Context depth is the other gap. Compared with Cursor's repo-indexed agent, Copilot's in-editor context is narrower on very large codebases in my experience, which shows up as more repeated explanations to the agent. And the whole product assumes a GitHub account and, for the enterprise features, GitHub Enterprise Cloud — Copilot is not available for GitHub Enterprise Server at all. Finally, model-specific data policies vary: read the Fable footnote in the docs before enabling those models in a regulated environment.

One detail worth knowing before a team rollout: because credits are pooled per user but governed by plan, a manager can predict roughly what a team consumes by watching one power user for a week. The billing overview page makes that visible per seat, which is the number I would put in a quarterly tooling budget rather than the sticker price.

Verdict

I rate GitHub Copilot 4.3. It earns the number on breadth — the widest IDE and surface coverage in this category, unmetered completions on every plan, and an agent that works where code review already happens — and on value, where the free tier and $10 Pro remain unusually generous for what they include. Usability is high because the tool is everywhere you work; the ecosystem is GitHub itself. The deductions are for the loss of fallback when credits run out, the annual-plan migration cliff, and narrower large-repo context than the specialist editors. For developers already living in the GitHub ecosystem, this is the default I would start with.

Sources

• https://docs.github.com/en/copilot/get-started/plans
• https://github.blog/news-insights/company-news/github-copilot-is-moving-to-usage-based-billing
• https://github.com/features/copilot/plans
• https://docs.github.com/en/copilot/concepts/billing-and-usage/individuals/billing
• https://docs.github.com/en/copilot/reference/ai-models/supported-models
• https://docs.github.com/en/copilot/reference/copilot-billing/models-and-pricing
• https://benchlm.ai/github-copilot/pricing
• https://shattered.io/github-copilot-pricing-ai-credits-claude-codex-agents-2026
• https://usagepricing.com/blueprint/github-copilot
• https://pecollective.com/tools/github-copilot-pricing