GitHub Copilot

What it is

GitHub Copilot is GitHub’s AI coding assistant, and in 2026 it is four products on one licence: inline completions and next-edit suggestions in VS Code, JetBrains, Visual Studio and Xcode; chat and agent mode in the editor; a cloud coding agent you assign an issue to and that returns a pull request; and Copilot code review on pull requests, plus the Copilot CLI. It is the incumbent — Microsoft reported 4.7 million paid Copilot subscribers in January 2026, up 75% in a year — and in most engineering organisations it is already bought before anyone in ops asks.

The thing to understand before reading any comparison written before June: Copilot changed its billing model. Since 1 June 2026 every plan is metered in GitHub AI Credits (1 credit = $0.01), charged by input, output and cached tokens at each model’s API rate. Completions and next-edit suggestions stay unlimited on paid plans and do not draw credits; chat, agent mode, the cloud agent, the CLI, Spaces and code review do.

Why ops teams run into it

GTM engineers, RevOps engineers and the one legal-ops or recruiting-ops person who writes Python usually get Copilot because the company’s GitHub organisation already has it. The question they bring is the one in the September bill: is a $19 Business seat still enough for agentic work — building an n8n node, a Salesforce sync script, an MCP server — or should the ops-dev group move to Cursor or Claude Code?

The answer depends on how much agent work you do, because that is what the credit cut hit. For June, July and August GitHub ran promotional included usage of $30 per Business seat and $70 per Enterprise seat. On 1 September that dropped to the standard $19 (1,900 credits) and $39 (3,900 credits). Teams that sized their habits on the summer allowance hit the pool limit in September.

Pricing reality

List prices, checked on GitHub’s plans page and docs today:

• Free — 2,000 completions and limited chat a month.
• Pro — $10/month; $10 base credits plus a variable “flex” allotment GitHub currently shows as $5.
• Pro+ — $39/month; $70 of credits shown including flex.
• Max — $100/month; $200 shown including flex. Added when individual sign-ups reopened in June after a pause that began 20 April.
• Business — $19/user/month, 1,900 credits pooled across the billing entity.
• Enterprise — $39/user/month, 3,900 credits pooled, adds GitHub.com chat with org-codebase indexing; requires GitHub Enterprise Cloud, which carries its own per-user fee.

Three billing details decide the real number. Credits are pooled: 10 Business seats share 19,000 credits, so two heavy agent users can drain the pool for eight completion-only users. Paid overage is on by default for organisations, with no automatic fallback to a cheaper model. And Copilot code review on private repositories also burns GitHub Actions minutes since 1 June.

A worked band for a 10-person ops-dev group on Business: $190 a month in seats. If the group keeps its summer usage — $30 of credits per head — it pays $110 a month in overage, an effective $30 per seat. That is still under Cursor Teams at $40 and well under a Claude Team premium seat at $100 annually; the cost case for leaving Copilot rarely holds on price alone. New organisations should note one more thing: GitHub paused self-serve Copilot Business sign-ups for Free- and Team-plan organisations on 22 April, routing new buyers to sales, and had not announced a reopening when we checked.

Best for

A RevOps engineer, GTM engineer or ops-adjacent developer in a company whose code already lives on GitHub, who wants completions all day and agent help a few times a week, and whose security team prefers a vendor already under contract. It wins on reach: one seat covers the IDE, the terminal, the pull request and the issue tracker, with org policy for which models are allowed.

Not for

An ops builder whose main mode is long agentic sessions — “read this repo, plan the integration, write and test it” — every day. That is the usage the credit model meters hardest, and it is where Claude Code and Cursor are built to spend tokens. Also not for a team that wants a stable model list: GitHub retired six models on 1 September and has six more scheduled for 19 October.

Versus the alternatives

• Cursor — the strongest challenger in the IDE, $20 individual and $40 per user on Teams. Pick it when the work happens inside the editor and you want the more capable agent experience and are willing to pay roughly double per seat.
• Claude Code, included in Claude Team seats ($20 standard, $100 premium, annual) — the fastest-growing tool in the segment. Pick it when the work is terminal-first, long-running agent sessions over a whole repo or MCP-connected systems.
• OpenAI Codex, bundled with ChatGPT Business — pick it when the company is standardised on OpenAI and you would rather not add another vendor.

Running Copilot for completions and adding Claude Code for agent work is a common and defensible split; see Claude vs Cursor and best AI coding assistants.

Watch-outs

• The pool drains silently, then the bill arrives. Overage is enabled by default and one user running the cloud agent all afternoon spends everyone’s credits. Guard: set user-level budgets for heavy agent users and an organisation budget in AI Controls before the first full month; if you want a hard stop, disable the “AI credits paid usage” policy.
• Model churn breaks pinned workflows. Retirements land every few weeks, and a prompt or custom agent tuned on one model shifts behaviour on its replacement. Guard: subscribe to the GitHub Copilot changelog, keep a short regression prompt set for any agent you depend on, and set the allowed models with the global model policy rather than per-user choice.
• AI review is not security review. A Wiz report that circulated in August showed its Red Agent finding a command injection in a GitHub Actions workflow in Snowflake’s public snowflake-connector-net repository that exposed Jira credentials; Snowflake fixed it and rotated the credential the same day. Whether Copilot wrote the vulnerable line is disputed — Wiz narrowed its claim to Copilot reviewing the pull request without flagging it. Guard: keep CodeQL or another static scanner and a human approver as required checks on any workflow file, and treat a Copilot review as one signal, never the gate.