On June 1, 2026, GitHub flipped the switch on one of the most consequential pricing changes in the history of AI coding tools. Copilot, used by millions of developers, moved away from its familiar flat-rate subscription and onto a metered model built around “GitHub AI Credits” that charge based on how many tokens you actually consume. The change had been announced back in late April by GitHub’s chief product officer, but the reality of living with it has landed hard across Reddit, X, and Hacker News.
What actually changed
The old system billed a low, predictable monthly fee tied to a quota of premium requests. The new system converts that monthly fee into an equivalent allowance of AI Credits, where one credit is worth one cent. Credits are then drawn down according to the input, output, and cached tokens processed by whichever model you invoke, priced at that model’s published API rate.
Base subscription prices did not change. Each paid tier now bundles a monthly credit allowance equal to its price: the $10 Pro plan includes $10 in credits, Pro+ includes $39, and the per-seat Business and Enterprise plans include $19 and $39 per user respectively. Annual subscribers keep the old model until their term runs out, after which they drop to the free plan unless they re-subscribe monthly.
What stays free
Importantly, the features that originally made Copilot popular are not metered. Inline code completions and Next Edit Suggestions remain unlimited and do not draw down credits at all. If you use Copilot mainly as a smart autocomplete, your day-to-day experience is essentially unchanged. The metering bites when you lean on the heavier, agentic capabilities: multi-step tasks, chat-driven refactors, and automated workflows that burn through tokens quickly.
Why developers pushed back
The backlash was immediate. Power users running agentic sessions projected cost increases anywhere from tenfold to fiftyfold, and individual reports circulated of monthly bills jumping from around $29 to several hundred dollars, with a few extreme projections running into the thousands. One developer on the top individual tier reported burning through roughly 8% of their monthly allowance in a couple of hours.
Two details amplified the frustration. First, the previous fallback safety net that let requests continue on a cheaper model was removed. Second, Copilot’s code review now runs on an agentic architecture hosted on GitHub Actions, so reviewing a pull request consumes both AI Credits and Actions minutes — a double charge for teams that automate reviews. To soften the landing, GitHub is handing Business and Enterprise customers temporary promotional credits through the summer, and it has introduced user-level budget controls alongside expanded context windows.
The bigger picture
None of this is happening in isolation. Usage-based pricing is the prevailing direction across the industry — OpenAI, Anthropic, and Google all bill API consumption by the token — and running large models at scale is genuinely expensive. GitHub frames the shift as the natural maturation of a product that has evolved from an autocomplete helper into a full agentic platform. For smaller teams and solo developers, though, the move trades predictability for precision, and predictability is exactly what made flat-rate pricing comfortable to budget around. Unsurprisingly, the change has sent some users to evaluate alternatives such as Cursor, Windsurf, and Claude Code.
The takeaway for anyone relying on AI coding tools is straightforward: understand your token economics. Know which features are free, which are metered, and roughly how your workflow translates into spend. The era of “set it and forget it” pricing for advanced AI assistance appears to be over.
Sources: TechCrunch, Visual Studio Magazine, gHacks, Dataconomy, and GitHub’s own community announcement.