GitHub Copilot

Rating reflects GitHub's documented plans and capabilities as of September 2026 plus our own sessions. Some vendor links on this site are affiliate links; they never affect a rating. See our disclosure and method.

What it does

Copilot started as autocomplete and is now four things under one login. Inline completions and next-edit suggestions finish the line you are typing. Copilot Chat answers questions about your code in the editor, on github.com and on mobile. Agent mode reads the repository, plans a multi-file change, edits, runs commands and iterates until the task is done. And the cloud agent does the same unattended: assign it an issue and it opens a pull request from a GitHub Actions sandbox. Copilot code review reads pull requests, and since the September 14 release it resolves its own comments when you address them, suggests commit messages and can run shell tools to validate a change.

Two surfaces arrived this year. The Copilot CLI brings agent sessions to the terminal on every plan, including Free. The Copilot app, generally available on July 7, 2026 for macOS, Windows and Linux, is a control center for running several agents in parallel, each in its own git worktree, with an Agent Merge feature that watches CI and reviewers until a pull request is ready. It also works with your own API key and no Copilot plan at all. Pro and above can additionally hand work to third-party agents, currently Claude Code and Codex, from inside GitHub.

Model choice is the other reason people pay. Free and Student get auto model selection only, which since September offers efficiency, balance and intelligence tiers. Pro unlocks a picker that, as of September 2026, spans GPT-6 Astra and the GPT-5.6 family, Claude Fable 5.1, Opus 5 and Sonnet 5, Gemini 3.8 Flash, Grok 4.6, Kimi K3 and Microsoft's own MAI-Code-1.1-Flash. Pro+ adds the premium tier including Opus; Max gets priority access to new models. Each model has a per-token price, so the choice is also a spending decision.

Where it wins

• Reach. No other card in the clubs suit covers this many surfaces: six editor families, the website, mobile, a terminal, a desktop app and pull-request review, on one subscription an employer probably already pays for.
• The free tier and the student plan. Two thousand completions plus limited agent use, vision on every plan since July, and a free Student plan for verified students. Teachers and maintainers of popular open source projects can get Pro free.
• Pro at $10. Unlimited completions, the cloud agent, code review, third-party agents and $15 of credits for $10. Pondero's July 2026 review called Pro+ the working tier for heavy users, but for most developers Pro is the plan.
• Models from everyone. Copilot is the easiest way to run Claude, GPT, Gemini, Grok and an open-weight Kimi model side by side without five API keys. Kimi K2.7 Code at $0.95 per million input tokens is a genuinely cheap way to burn fewer credits on routine work.
• Admin control. Business and Enterprise pool credits across the organization, set per-team cost centers, let users request more budget when they run out, exclude files from Copilot and block suggestions that match public code.

Where it loses

• Training on by default. From April 24, 2026, interaction data from Free, Pro and Pro+ users trains GitHub's models unless you opt out. That includes the code around your cursor in private repositories while you are actively using Copilot. Cursor asks you to turn Privacy Mode on; Copilot asks you to turn training off.
• Credits are harder to predict than requests. Before June 1 you counted premium requests. Now a long cloud-agent session on Claude Opus 5 at $25 per million output tokens can cost more than a day of chat on GPT-5.6 Luna at $1.20. The dashboard shows spend, but the meter runs while you think.
• Flex can shrink. Of Pro's $15 in credits, $10 is guaranteed and $5 is a flex allotment GitHub says will vary over time. Pro+ is $39 guaranteed of $70 and Max $100 of $200. Budget on the base, not the total.
• Multi-file editing trails Cursor. Two independent reviews this year, Pondero in July and AIUnpacking in May, rated Copilot highly and still said Cursor's Composer edits large multi-file changes more cleanly. Agent output needs review; it produces plausible wrong refactors.
• Anthropic's Fable models keep data. GitHub's own docs note that when Claude Fable 5 or 5.1 is used, Anthropic retains prompts and outputs by default for safety classifiers, with a zero-retention exemption for approved enterprises only through the end of 2026. Other Claude models run under zero retention.
• Students lost the premium models. The March 13, 2026 move to a dedicated Student plan removed premium model access and third-party agents from the education benefit.

Plans and pricing, September 2026

One AI credit is one cent. Usage beyond the included allowance is billed at the same per-token rates, and Business and Enterprise pay one cent per credit past the pool. Completions never consume credits. Subscribers on legacy annual Pro and Pro+ plans still use the old request-based multipliers, and anyone who subscribed through the iOS or Android app cannot buy extra credits.

Score breakdown

Alternatives

• Cursor if you want the editor built around agents, with cleaner multi-file edits and an opt-in Privacy Mode.
• Claude if you want Claude Code on its own terms, with Anthropic's plans and data controls rather than GitHub's.
• ChatGPT if you want Codex bundled with a general assistant and a free tier that includes it.

Quick answers

Is GitHub Copilot free?

There is a Free plan with 2,000 code completions a month, limited chat and agent use on auto model selection, the Copilot CLI and the Copilot app. Verified students get the separate Copilot Student plan at no cost. Pro is $10 a month and includes $15 worth of AI credits plus unlimited completions.

What are GitHub AI credits?

Since June 1, 2026, Copilot bills chat, agents, code review and the cloud agent by the token, converted into AI credits at one cent each. Pro includes 1,500 credits a month ($15), Pro+ 7,000 ($70) and Max 20,000 ($200), split between base credits that match your subscription price and a flex allotment GitHub can change. Completions stay unlimited on paid plans, and you can buy more credits at a cent each.

Does GitHub Copilot train on my code?

On Free, Pro and Pro+, yes by default since April 24, 2026: prompts, accepted suggestions, code context around your cursor and file names can be used to train GitHub's models unless you opt out under Settings, Privacy. Business and Enterprise plans are excluded. Data may be shared with Microsoft affiliates but not with third-party model providers.

More cards

• 
  Tool cards

  Cursor

  The agent-first editor, rated and priced.

• 
  Tool cards

  Claude

  The writer's model, rated and priced.

• 
  Guides

  Build your first AI stack

  Which cards to hold together, and what it costs a month.