GitHub Quietly Made Copilot Code Review Burn Two Budgets at Once
GitHub says prices arent changing. Check your bill.
On June 1, GitHub Copilot switched from a flat monthly fee to a token-based billing system. The base subscription tiers look the same: Pro is still $10 a month, Business is still $19 per user. But the way those dollars translate into actual AI access has fundamentally changed, and for a significant number of developers, the difference is not subtle.
The pricing GitHub published shows the math clearly. GPT-5.2, one of the more capable models available in Copilot, costs $1.75 per million input tokens and $14 per million output tokens. Claude Opus 4.8 runs $5 input and $25 output per million tokens. At those rates, a single long coding session can burn through what used to be a weeks worth of flat-rate allocation in minutes.
Developers are posting the results. One Reddit user reported their Copilot costs jumping from roughly $29 a month to around $750. Another showed a bill climbing from $50 to $3,000. These are not typical users: they are the power users GitHub explicitly built agentic features to serve. But the numbers are real, and they are landing on desks right now.
There is a second cost layer that GitHub disclosed less prominently. Copilot code review, the feature that analyzes pull requests, runs on GitHub Actions infrastructure. Under the new billing, each code review pulls from two separate meters simultaneously: the AI Credits that fuel the model, and the Actions minutes that power the underlying compute. GitHub confirmed this in its community discussion forum and documentation. It did not lead with it. The result is a compounding charge that organizations will see when their next CI/CD bill arrives, not when they read the migration blog.
GitHub announced the transition in April, framing it as a correction: the flat-rate model was a loss leader, and the inference costs of agentic coding sessions bear no resemblance to a quick autocomplete. Usage-based pricing is more honest. You get what you pay for.
You also pay for what you get. For years, Microsoft and GitHub actively encouraged developers to use Copilot heavily: autonomous agents, multi-step refactors, full repository analysis. The product roadmap leaned into scale. The pricing model is now penalizing exactly the usage pattern the company promoted. That is not a subtle contradiction.
The impact falls unevenly. Enterprises with large developer budgets will adjust. Small teams and independent contractors face the sharpest cost pressure. This is the cloud computing story again: metered infrastructure always looks fine until you get the bill.
The strongest counterargument is that the flat-rate era was always temporary, and that metered pricing gives engineering leaders the visibility they need to actually manage AI tool spend. GitHub offers a discount for auto-selection and publishes the full pricing table. Sophisticated teams can optimize. They can also leave: Cursor, JetBrains AI, and other Copilot alternatives sell annual plans that look more predictable by comparison.
Microsoft did not respond to a request for comment from TechCrunch.
What GitHub will not do is return to the old model. The transition is complete. The credits are burning.