5-Min Brief: GitHub Just Changed How It Charges for AI — and Developers Are Furious.

What you need to know — in 30 seconds

• GitHub Copilot — the AI coding assistant used by 4.7 million paying developers — switched to token-based billing on June 1
• The flat monthly subscription ($10-$39/month) is gone, replaced by a system that charges for every word the AI reads and writes
• Developers running heavy AI sessions are reporting cost increases of 10x to 50x — monthly bills jumping from $29 to $750, or from $50 to $3,000
• Basic autocomplete remains free — the pain falls on developers using the new agentic AI features Microsoft has been aggressively promoting

Yesterday we covered Microsoft's Build conference and its big bet on AI agents — software that handles entire workflows autonomously. The same day those announcements were happening on stage, 4.7 million GitHub Copilot subscribers were discovering what powering those agents actually costs. The reaction has been swift, loud, and mostly furious.

What changed on June 1

For most of its existence, GitHub Copilot worked like a gym membership. You paid a flat monthly fee — $10 for individuals, $19 for businesses, $39 for enterprise — and you could use it as much as you wanted. One price, unlimited usage.

That model is gone.

Starting June 1, Copilot runs on what GitHub calls "AI Credits" — a token-based system that charges for every piece of text the AI processes. Input tokens (what you send in — your code, your questions, the files you share), output tokens (what the AI sends back), and cached tokens (context the model has already seen) all count against your credit balance.

When you run out of credits, premium features stop working until next month. If you want more, you pay more.

GitHub's own blog post announcing the change, published in late April, was candid about why: "It had become common for a handful of requests to incur costs exceeding the plan price." In other words, the heaviest users were costing GitHub more than they were paying. The flat rate had become unsustainable.

Who this hurts — and who it doesn't

This is the detail most coverage is getting wrong: the change doesn't affect everyone equally.

Code completions — the inline autocomplete that appears as you type — remain included in all plans and do not consume AI Credits. If you use Copilot primarily to suggest the next line of code as you write, your bill doesn't change. At all.

The pain falls on developers using Copilot's newer agentic features — the ones Microsoft has been promoting most aggressively. When an AI agent reads through your entire codebase to understand context, generates hundreds of lines of code, iterates on feedback, and runs tests — every one of those operations consumes tokens. A session that costs $0 under the old model might cost $20 or $50 under the new one.

For individual developers using Copilot casually, this is probably fine or even an improvement. For engineering teams running intensive AI coding sessions against large codebases — the exact use case Microsoft built its Build conference around — the cost structure has changed fundamentally.

The numbers people are sharing

Across Reddit, X, and GitHub's own community forums, developers have been sharing their projected costs since the billing switch. The numbers are striking.

One developer reported their monthly bill jumping from $29 to $750. Another cited projections of $50 becoming $3,000. TechCrunch, covering the backlash, called it the end of "the golden age" of GitHub Copilot. The GitHub community discussion thread has hundreds of comments, most negative.

Microsoft has offered some mitigation: Business plans receive an additional $30 per user per month in promotional credits through August 2026, and Enterprise plans receive an additional $70 per user per month through the same period. After September, those promotions end.

Organization administrators can set budget caps to prevent runaway costs at the user, team, and enterprise level. Business and Enterprise plans have also gained pooled credit usage, so unused credits from lighter users can offset heavy users within the same organization — a structural improvement over the old model, though it doesn't reduce the underlying costs.

Why Microsoft did this — the honest version

Microsoft's official framing is that token-based billing aligns pricing with actual usage and creates a more sustainable business. That's true as far as it goes.

But there's a harder truth underneath it. Internal documents obtained by journalist Ed Zitron showed that the week-over-week cost of running GitHub Copilot had nearly doubled since January 2026, making the transition more urgent than a planned product strategy would suggest.

The agentic AI features Microsoft showcased at Build yesterday — AI that reads entire codebases, writes and tests code autonomously, manages pull requests — are extraordinarily compute-intensive. Under the old flat-rate model, those features were essentially subsidized by the majority of lighter users. Token-based billing ends that subsidy and makes every user pay for what they actually consume.

This is the underlying economics of AI in sharp relief. The powerful new capabilities — the ones that make AI genuinely transformative rather than just helpful — cost real money to run. The industry built its user base on flat rates and free tiers. Now, as those capabilities become more powerful and compute costs compound, someone has to pay.

This week, that someone is the developers who embraced agentic AI the fastest.

What this means for the broader AI industry

GitHub Copilot is the largest AI developer tool in the world. How it prices its product matters beyond just the developers affected today.

The token billing model is almost certainly where the rest of the industry is heading. Every AI tool running on frontier models has the same underlying cost structure — the more the AI does, the more it costs. Flat rates work when usage is light and predictable. They break down when AI agents start running hour-long sessions against million-line codebases.

What's happening with Copilot this week is a preview of conversations that will happen across every AI productivity tool in the next 12-24 months: how much does this actually cost to run, who pays for it, and what's the right model for pricing AI that genuinely does work rather than just answering questions?

Nobody has a clean answer yet. GitHub just became the first major platform to find out what happens when you tell power users the party is over.

HumanReadable-AI covers AI news in plain English every weekday. Subscribe below — free, no jargon, always under five minutes.