On June 1, 2026, GitHub made a quiet change. Most of the internet focused on developer wallets. But the real story is about product economics — and it affects a lot more than Copilot users.
Here’s what actually happened, and why it matters.
Flat-Rate AI Is Gone
GitHub Copilot used to cost a flat monthly fee. Ten dollars. Thirty-nine dollars. Predictable. You paid it and forgot about it.
That model is dead now.
GitHub announced that all Copilot plans would transition to usage-based billing on June 1, 2026. Instead of counting premium requests, every plan now includes a monthly allotment of GitHub AI Credits, with the option for paid plans to purchase additional usage.
One AI Credit costs one cent. Your plan comes with a monthly bucket of them. Usage is calculated based on token consumption — including input, output, and cached tokens — using the published API rates for each model.
A token, simply put, is a small chunk of text. Every word your team sends to the AI and every word it sends back gets counted and charged.
Code completions and Next Edit suggestions remain included in all plans and don’t consume AI Credits. For developers doing basic work, the impact is minimal. But for teams running agentic workflows — where AI works autonomously across multiple files, writes code, reviews it, loops through tasks on its own — costs are jumping 10x to 50x. One user reported burning through 82% of their monthly credits on day one.
The community announcement thread collected nearly 900 downvotes. People feel burned. That reaction is understandable.
Why People Are Angry
Microsoft spent three years encouraging developers to lean into AI. Use agent mode. Build workflows around Copilot. Let it run.
Developers did exactly that.
Then the pricing changed. The workflows GitHub encouraged — the ones teams built their productivity around — are now the most expensive ones under the new model. That’s a trust problem, not just a cost problem.
GitHub’s explanation, written by Mario Rodriguez on the official blog, isn’t unreasonable. GitHub framed the change as aligning Copilot pricing with actual usage — describing it as “an important step toward a sustainable, reliable Copilot business and experience for all users.”
There’s also a meaningful structural difference in the new model. Previously, users who exhausted their premium requests could fall back to a lower-cost model and continue working. Under the new model, usage is governed by available credits and admin budget controls — and that fallback is gone.
Whether you see that as fairer pricing or a harder wall depends on which side of it you’re on.
The Bigger Problem No One Is Talking About
Here’s where most coverage misses the point.
Token-based billing isn’t just GitHub’s new model. It’s the direction the entire AI industry is moving. OpenAI, Anthropic, Google — every major API already charges by the token. GitHub just made it visible to a much wider audience.
And if you’re building software products with AI inside them, this creates a cost exposure on both sides.
Your engineering team pays per token to build. Your customers trigger AI costs every time they use what you shipped. Two meters running simultaneously. Neither stops when you’re not watching.
Most SaaS businesses price by seat or by tier. Costs are relatively fixed. AI breaks that model completely. A poorly designed AI feature doesn’t just cost money upfront — it gets more expensive with every new user you add.
That’s not a software problem. That’s a unit economics problem.
What a Bad AI Feature Actually Looks Like in Production
Here’s a concrete example worth thinking through.
A team ships a document summary feature. Users love it. Engagement is high. The product team celebrates. But nobody checked the cost design.
Every time a user clicks summarize, the product sends the entire document to the most powerful model available. No caching — the same context gets re-sent every single time. No model routing — a task a smaller, cheaper model could handle goes to the most expensive one. No spending cap — the cost scales silently with every new customer.
At a hundred users, nobody notices. At ten thousand, the company is losing money on its most popular feature.
This isn’t theoretical. It’s a pattern that shows up repeatedly when AI integrations are reviewed in production. The demo never reveals it. The invoice does.
How Engineers Should Actually Think About This
The teams handling this well treat AI cost as an engineering constraint — something you design around from the start, not a finance problem you discover at quarter-end.
In practice, that means a few specific things.
Set token budgets per feature before you build. Know what each user interaction is expected to cost before you ship. If you can’t answer that question, you’re not ready to ship.
Route by task complexity. Simple tasks — autocomplete, classification, basic formatting — go to smaller, cheaper models. Hard reasoning and generation tasks go to the frontier models that justify the cost. The model-price spread across GitHub’s own menu is enormous — GPT-5.5 output is 24x the cost of GPT-5.4 nano output. A careless model selection is now effectively a billing bug.
Cache repeated context. If your feature sends the same background information on every call, you’re paying for it every time. That’s not complex engineering — it’s just not wasting money.
Monitor AI spend like you monitor errors. Set alerts. Know your cost-per-user baseline. Flag anomalies in real time, not monthly when the bill arrives.
None of this is advanced. But it requires someone asking the cost question before the feature ships.
What This Means If You’re Not an Engineer
You don’t need to understand token economics at a technical level. But if AI is anywhere in your product roadmap, you need to be asking the right questions.
What does this feature cost per user? Does that cost scale with usage? What happens to margins at ten times current usage? Who’s monitoring AI spend — and how often?
If your team can’t answer those clearly, that’s worth knowing now. AI features that users love but that quietly destroy margins aren’t features. They’re liabilities with a good interface.
The Honest Bottom Line
GitHub’s pricing change is frustrating for developers. But it’s also clarifying.
The era of treating AI as a flat, predictable cost is over. Token billing is now the default — across the tools your team uses and the products you ship. That’s not reversing.
GitHub’s changelog confirmed that as of June 1, usage-based billing is now live for all users — and Copilot code review now consumes GitHub Actions minutes in addition to AI Credits. The scope of what gets metered is already expanding.
The teams that adapt fastest are the ones who stop thinking of AI cost as something that happens to them and start designing around it deliberately.
The meter is already running. The question is whether you’re watching it.
Sources: GitHub Blog (Mario Rodriguez, April 27, 2026), GitHub Changelog (June 1, 2026), GitHub Docs — Usage-Based Billing, Microsoft Partner page – GitHub Usage-Based Billing, TokenMix Research Lab, GitHub Community Discussion #192948, GitHub Community Discussion #197089
