GitHub Copilot flipped its billing model on June 1, moving from a flat monthly subscription to token-based consumption pricing. The result: developers are waking up to projected costs that are, in some cases, orders of magnitude higher than what they were paying before.
What Actually Changed
Until now, Copilot operated on a predictable flat-rate subscription. Token-based billing means you pay for what you use, and "what you use" turns out to be a lot more than most developers assumed. Every completion, every agentic loop, every multi-file edit burns tokens. The more Copilot does on your behalf, the more you pay.
According to TechCrench's coverage, Microsoft did not respond to requests for comment on the pricing change. No official breakdown of per-token rates has been published, which makes it nearly impossible for developers to forecast costs before they hit.
The Numbers Are Ugly
User reports on Reddit and X tell the story bluntly. One developer posted that their bill would jump from $29/month to nearly $750/month. Another shared a screenshot showing costs climbing from around $50 to roughly $3,000. Those aren't rounding errors. They're a different order of magnitude entirely.
"What a joke," one Redditor wrote, describing their projected cost increase from $29 to nearly $750 a month.
These figures come from user reports and billing projections, not a Microsoft-published pricing calculator. Specific per-token rates haven't been confirmed publicly, so treat the dollar amounts as indicative rather than definitive. But even if the real numbers land somewhere in the middle, the direction is unmistakable.
Why This Stings More Than Usual
The frustration isn't just sticker shock. Microsoft spent the past year actively making it easier to burn through tokens. Copilot's agent features, extended reasoning, and multi-step code generation workflows were all pushed as selling points. Users built habits and entire workflows around those capabilities under flat-rate pricing.
"Microsoft provided this billing method and they kept making it easier and easier to burn through massive numbers of tokens on single premium requests that could churn for hours or even days while spawning dozens or even hundreds of sub-agents."
The product was designed to encourage heavy usage, and now heavy usage is expensive. That's a legitimate grievance, not just complaining.
The Counterargument
Not everyone is sympathetic. Some developers pushed back in the same threads, arguing that sky-high bills point to inefficient workflows rather than predatory pricing.
"The only way it gets crazy like that is if you are purely 'vibe coding' with a ton of bloated iterations."
There's also a more cynical read, captured in one Reddit comment that's hard to argue with: "Holy fuck how much money was Copilot losing." Flat-rate AI tooling at scale was never a sustainable model for Microsoft, and token pricing is the correction. That doesn't make it less painful for users, but it explains why the change happened.
Where Developers Are Looking Now
The timing is notable. The AI coding tool market isn't short on alternatives. Tools like Cursor, JetBrains AI Assistant, and direct API access to models like Claude are all in the conversation. Whether any of them offer better value at high usage levels is a comparison the available data doesn't fully resolve, but the developer community is clearly in comparison-shopping mode right now.
If you're a heavy Copilot user, the most important thing you can do before the billing hits is audit how you actually use it. Agent features and long-running agentic loops are almost certainly where token costs spike. Knowing your usage patterns matters a lot more now than it did a month ago.
Bottom Line
GitHub Copilot's token billing switch is a real problem for developers who relied on flat-rate predictability, especially those who leaned into the agent features Microsoft itself promoted. Until Microsoft publishes a clear pricing breakdown and usage calculator, it's difficult to make an informed decision about staying on the platform. Light users doing standard completions may find the math still works. If you're running agentic workflows or heavy multi-file edits, run the numbers before the June bills arrive.