As of today, June 1, 2026, GitHub Copilot no longer sells you unlimited AI. It sells you a monthly credit allotment, and when you blow past it, you pay overage. This is not a small footnote update. For developers doing heavy agentic work, this is a fundamental repricing of how you use the tool.
Reports are already coming in from developer forums: bills jumping from $29/month to $750. From $50 to $3,000. Those aren't outliers — they're power users who spent the last year treating Copilot's flat rate as an all-you-can-eat ticket for agentic sessions. That ticket has been torn up.
What Actually Changed
The plan prices are unchanged. Pro is still $10/month. Pro+ is still $39/month. Business is $19/user, Enterprise $39/user. GitHub isn't raising the headline number — they're changing what you get for it.
Each plan now comes with a monthly AI Credit allotment. Use within that allotment and nothing changes. Go over, and you're paying overage rates. The model that always burned credits hard but was previously throttled by fallback — that safety valve is now gone.
The thing that's quietly disappeared
Previously, when you exhausted your premium request units (PRUs), Copilot fell back to cheaper models. That kept bills predictable and agentic sessions affordable. That fallback mechanism no longer exists. When you hit your limit, you're paying full overage rates, full stop.
What Burns Credits and What Doesn't
Not everything costs you credits. Code completions — the inline suggestions as you type — are still free. Next Edit Suggestions are free. The features that work like glorified autocomplete don't touch your allotment.
What burns credits hard: Copilot Chat, large context operations, automated code review, and agentic sessions. That last one is the killer for heavy users. Agentic sessions — where Copilot is autonomously planning, executing, and iterating across your codebase — are routinely consuming $30 to $40 per session for Pro users alone.
- Code completions: free, no credits consumed
- Next Edit Suggestions: free, no credits consumed
- Copilot Chat: burns credits
- Large context operations: burns credits
- Automated code review: burns credits
- Agentic sessions: burns credits hard — $30–$40 per session is being reported
Who Gets Hit Hardest
The developers in pain right now are the ones who treated Copilot's flat-rate pricing as a structural advantage. They picked Copilot specifically because it let them run long agentic sessions — the kind that Cursor or Claude Code would charge serious money for — under a $39/month ceiling. That calculus is now completely inverted.
If you've been using agentic sessions daily, you need to do the maths today. Not tomorrow. Run a single session, watch what it consumes, and project it across a month. The sticker shock is real and it's happening to people who thought they had a sustainable workflow.
“Copilot's flat-rate agentic pricing was genuinely good value. It was also probably unsustainable. GitHub has decided that now is the time to find out.”
Light Users: Relax
If Copilot is part of your workflow but not the centre of it — you use autocomplete daily, open Chat a few times a week for quick questions, and rarely run anything you'd describe as an agentic session — you're probably fine. Your monthly usage will likely stay within the allotment and your bill stays flat.
The key is to actually check. GitHub is rolling out a usage dashboard alongside this change. Look at where your credits are actually going before you panic. If Chat and completions are your primary use cases, this change might not touch you at all.
What To Do If You're a Power User
First: instrument your usage immediately. You cannot manage what you cannot see. Check the GitHub dashboard, understand where your credits are being consumed, and identify the sessions that are costing the most per unit of value.
- Audit your agentic sessions. Not all of them are equally valuable. The expensive ones that output mediocre results are now especially expensive.
- Consider the alternatives seriously. Claude Code Pro is $100/month with generous agent usage. Cursor is competitive on agentic pricing. If you were using Copilot specifically for agentic work at flat rate, that specific use case needs re-evaluation.
- Set a budget cap immediately. GitHub allows you to configure a monthly spend limit. Do this today before your next session, not after your first bill lands.
- Be more deliberate with context window usage. Large context burns more credits. Tighter, more focused prompts are now cheaper to run.
The Honest Take on What This Means
GitHub building Copilot for Microsoft means this pricing model has been approved at the very top. This is not a beta experiment. Usage-based billing for AI is the direction the entire industry is heading, and Copilot flipping the switch today is the clearest signal yet that the "unlimited AI for a flat fee" era is over.
That's not inherently bad. Usage-based pricing aligns incentives correctly — you pay for value you consume rather than subsidising heavy users through a flat pool. But it does require you to actually think about your AI usage as a cost centre, which most developers haven't had to do yet.
The developers who adapt fastest are the ones who treat their AI tooling stack like the infrastructure decision it now is: deliberate, monitored, and chosen based on what delivers value at what cost. Welcome to the new normal.