From flat subscriptions to usage-based GitHub AI Credits
On June 1, GitHub Copilot pricing quietly changes in a major way. The familiar flat subscriptions stay — Copilot Pro, Pro+, Business, and Enterprise keep their listed monthly prices — but the hidden meter behind them is being replaced with usage based billing driven by tokens instead of abstract “premium requests.” Every paid plan will now come with a monthly allowance of GitHub AI Credits, calculated against real token consumption for each model, including input, output, and cached tokens. Once your Copilot AI Credits run out, higher-end features simply stop working until the next monthly reset or until you buy more usage. GitHub argues that this shift better aligns GitHub Copilot pricing with the actual cost of running today’s agent-style, long-running sessions, where a single multi-hour Copilot Workspace task can cost orders of magnitude more to serve than a quick inline suggestion.
How Copilot AI Credits are actually consumed across features
Under the new Copilot token model, not everything you do in your editor will eat into your GitHub AI Credits. GitHub has confirmed that standard in-editor code completions and Next Edit suggestions remain included in all plans and do not consume credits, which is good news if you mainly rely on Copilot for autocomplete-style assistance. Credits are instead spent on heavier features: multi-step chat sessions, repository-wide refactors, agentic workflows, and Copilot code review. Behind the scenes, each request is converted into token usage based on the model’s published API rates, covering your prompt, Copilot’s response, and cached context tokens. The more context you feed in — big files, whole repos, long chats — and the more capable the model, the faster your Copilot AI Credits will drain. Unused credits do not roll over, so any remaining balance is wiped at the monthly reset.
No more free fallback models: what happens when credits hit zero
One of the most developer-visible changes is the end of free fallback models. Previously, when you exhausted your premium request allowance, Copilot would quietly fall back to a cheaper model so you could keep working, just with less capable AI. Under usage based billing, that safety net disappears. When you run out of Copilot AI Credits, all credit-consuming features pause until your next monthly allotment lands or you top up with additional usage; only core completions and Next Edit keep functioning. For individuals, that means heavy experimentation with advanced chat or long-running agents can abruptly cut off mid-month. For small teams, a few power users can burn through shared credits faster than expected. GitHub is rolling out a preview bill experience so users and admins can see projected costs ahead of time, but day-to-day, you will need to actively watch balances instead of assuming Copilot will gracefully downgrade.
Who pays more, who pays less, and how code review hits CI budgets
The practical cost impact depends on how you use Copilot. Light and moderate users — mostly relying on inline completions, occasional chat, and small refactors — are likely to see similar or slightly better value, because those low-intensity interactions barely touch their Copilot AI Credits. Heavy users, especially those running long agentic sessions, whole-repo transformations, or high-end models, will almost certainly pay more once their included credits are gone. Teams using Copilot code review face an extra wrinkle: each review now consumes both Copilot AI Credits and GitHub Actions minutes when run on private repositories, with any overage billed at standard GitHub Actions rates. Public repos remain exempt from Actions charges. For small teams, that effectively ties Copilot code review cost to existing CI/CD budgets, forcing leads to think about how many automated reviews they can afford alongside their test and deployment workflows.
Controlling usage based billing: practical tactics and competitor context
For everyday developers and team leads, the key to extracting value from the new Copilot pricing is intentional usage. Start by watching the preview bill and monthly Copilot AI Credit reports to spot which workflows burn the most tokens. Set internal guidelines: prefer inline completions where possible, use chat for targeted questions instead of sprawling sessions, and reserve agentic multi-hour tasks for genuinely high-impact work. For Copilot code review, consider running AI review only on critical branches or large, complex pull requests, and pairing it with human review rather than replacing it. Competitors such as Anthropic’s Claude Code, Cursor, and other AI coding assistants already lean on token-based, usage based billing, often with clearer per-model pricing and fewer hidden meters. GitHub’s move closes that gap but also fuels concerns that this is a stealth price hike on power users, even as the company maintains that it simply reflects real inference costs.
