GitHub Copilot's new billing model is turning AI coding from a flat subscription into a visible token meter. Developers on Reddit reported jumps from $50 to $3,000 a month, and from $29 to $750, after Copilot moved toward usage-based AI Credits.
The subscription prices are still familiar: Pro, Pro+, Business and Enterprise include monthly credit allotments. The difference is that the included amount is no longer a hard ceiling. Once credits are exhausted, heavy usage can continue to generate charges.
Agents change the economics
Basic completion and Next Edit remain included, but agent mode is different. When Copilot works across a repository, loops through changes and uses large contexts, input, output and cached tokens can accumulate quickly.
Some developers blame inefficient vibe coding and bloated iterations. Others argue Microsoft and GitHub encouraged broad AI use and then moved the bill to users. GitHub's shift makes the underlying inference cost visible after two years of subsidized AI coding tools.
A wider pricing signal
The issue is unlikely to stay with Copilot. Cursor and other AI IDEs face the same arithmetic: one agent-heavy user can consume far more compute than many autocomplete-only users. A flat monthly price becomes harder to defend as agent workflows expand.
The practical lesson for developers is simple but uncomfortable. AI coding now has a budget line. Cheap, powerful and unlimited agent work may not coexist for long.
Sources: The GitHub Blog, TechCrunch, Reddit user reports cited in coverage, CocoLoop; checked the AI Credits change, plan credit amounts, transition allowances and reported $50-to-$3,000 and $29-to-$750 bill jumps.