GitHub Copilot is about to charge developers very differently, and a lot of them are furious. According to TechCrunch AI, Microsoft is scrapping Copilot’s flat subscription model in favor of token-based usage billing starting June 1. The short version: instead of paying a low monthly rate tied to requests, you’ll pay for how many tokens you burn while you work. For some users, that math looks brutal.
What’s changing
The old model was simple. A flat fee, a request-based cap, and predictable monthly costs. The new model ties your bill directly to consumption. The more tokens your prompts and agent runs chew through, the more you pay.
That’s a fundamental shift in how Copilot’s economics work, and it lands hardest on solo developers and small teams. Big enterprises can absorb variable costs. A freelancer budgeting month to month cannot.
Why developers are upset
TechCrunch AI collected a wave of reactions from Reddit and X, and the numbers people are sharing are eye-watering:
- One Redditor said their bill would jump from roughly $29 a month to nearly $750. “What a joke,” they wrote, adding they’re cancelling because “it is no longer cost-effective or useful in any practical way.”
- Another shared a screenshot appearing to show costs climbing from around $50 to about $3,000. “WOW, didn’t expect new pricing model to be this ridiculous,” they posted.
Those are extreme cases. But they spread fast, and they capture the core fear: a tool people built into their daily workflow could suddenly cost 10x or more.
The other side of the argument
Not everyone is sympathetic. Some Copilot users pushed back hard, arguing the giant bills come from sloppy usage, not the pricing itself.
One user said they work all day and “barely” hit overage charges, and they doubt the horror stories reflect real workload complexity. Their take: “The only way it gets crazy like that is if you are purely ‘vibe coding’ with a ton of bloated iterations.” Used as a disciplined tool, they argue, Copilot stays affordable even for small shops.
Then there’s the question nobody at Microsoft is answering out loud. “Holy fuck how much money was copilot losing,” one Redditor asked. It’s a fair point. The flat-rate model almost certainly subsidized heavy users, and token billing looks a lot like Microsoft trying to stop the bleeding.
The fairness problem
Here’s what stands out to me. Even if some bills are inflated by careless use, Microsoft built the system that encouraged that behavior in the first place.
One user put it sharply, noting Microsoft “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.” You can’t push users toward heavy agent workflows and then act surprised when they rack up heavy bills. TechCrunch AI reports Microsoft did not respond to a request for comment before publication.
Why this matters
This is bigger than one product’s price list. Copilot was the tool that normalized AI-assisted coding for millions of developers. Its pricing set expectations for the whole category.
The move to usage-based billing signals where the industry is heading. The era of flat-rate, all-you-can-eat AI tools looks like it’s ending as providers confront the real cost of inference. Expect more vendors to follow, especially for agentic features that spin up long-running tasks and sub-agents behind the scenes.
What to do now
If you rely on Copilot, a few practical moves before June 1:
- Check your current token consumption so you can estimate your new bill instead of guessing.
- Tighten your workflow. Fewer bloated iterations and runaway agent runs directly lower your cost.
- Compare alternatives. Rivals on different pricing models may be cheaper for your specific usage pattern.
- Set a budget alert if Microsoft offers one, so a single expensive run doesn’t blindside you.
The practical upside of usage billing is that disciplined developers may actually pay less. The risk is that experimentation gets expensive, and that’s exactly what made these tools fun to learn on. You can find the full breakdown and developer reactions at the original TechCrunch AI report.