On the night of June 1, one developer opened their GitHub billing dashboard and looked at the number twice.
$29/month → $750. A single billing cycle was all it took.
Another developer got $3,000. GitHub Copilot switched to AI Credits metered billing, and the developer backlash was immediate.
What actually changed?
As of June 1, 2026, GitHub Copilot overhauled its pricing model. The new system revolves around AI Credits.
1 AI Credit = $0.01 USD. Chat, agent mode, and Copilot CLI all consume credits. Code completions and inline suggestions stay unlimited on paid plans — but the moment you open a chat or spin up an agent, the meter starts.
| Plan | Monthly fee | Included AI Credits | Overage rate |
|---|---|---|---|
| Free | $0 | Limited | Blocked |
| Pro | $10 | ~$15 worth | $0.01/credit |
| Pro+ | $39 | ~$70 worth | $0.01/credit |
| Max | $100 | ~$200 worth | $0.01/credit |
The Pro plan ($10) looks reasonable with $15 in credits — until you see how fast agent mode burns through them.
Why are some devs getting $750+ bills?
The developers hit hardest have one thing in common: they use agent mode heavily. When an agent gets a prompt like "refactor this," here's what actually happens under the hood:
- Context gathering
The agent reads 10–20 relevant files. Each file access consumes tokens. - Planning
It internally reasons about what to change and how. That reasoning loop costs credits. - Code execution
Changes are applied file by file. Each write is a round trip. - Verification loop
"There's an error, let me fix it" — this can loop indefinitely. Each cycle burns more credits.
One "agent run" can translate to hundreds of individual API requests.
The most dangerous usage pattern
It's called "vibe coding" — giving the AI a vague direction and letting it loop until something works. Without a tight scope, the agent keeps iterating and credits stack up exponentially.
A lot of the backlash landed on Microsoft. "They kept making it easier and easier to burn through massive numbers of tokens, and now they're billing for it" — that sentiment flooded Reddit and X. TechCrunch put it bluntly: "The golden age of GitHub Copilot is at an end — for the little guy, at least."
Stay with Copilot or switch?
If you primarily use inline code completions, this change barely affects you — those remain unlimited on paid plans. The hit is concentrated on devs who rely on agent mode and chat.
| GitHub Copilot Pro | Cursor Individual | |
|---|---|---|
| Monthly cost | $10 + variable overage | $20 flat rate |
| Agent usage cost | Credits consumed | Included in flat rate |
| Cost predictability | Unpredictable | Predictable |
| Code completions | Unlimited | Unlimited |
For agent-heavy workflows, Cursor ($20 flat) or Amazon Q Developer Pro ($19 flat) offer predictable pricing with no surprise bills.
What you can do right now
- Check your usage first
GitHub Settings → Billing → Copilot shows your credit breakdown. Find out your agent/chat vs. completions ratio before making any moves. - Turn off paid overages
In Copilot settings, set paid overages to OFF. This blocks charges beyond your included credits — your first line of defense against bill shock. - Narrow your agent scope
Instead of "refactor the whole codebase," try "add types to this one function in src/auth.js." Smaller context = fewer agent loops = fewer credits. - Swap chat for completions
For simple suggestions, reach for inline autocomplete instead of chat. Completions are free on paid plans. - Trial an alternative for 30 days
If you run agents daily, try Cursor or Amazon Q for a month. Both have free tiers to start.
Want to go deeper?
TechCrunch original reporting Developer reactions and real billing case examples. techcrunch.com
GitHub Copilot official plans page Compare plans and included credits side by side. github.com
Cursor pricing The most-compared alternative for agent-heavy developers. cursor.com
Amazon Q Developer pricing $19/month flat-rate AI coding tool for AWS ecosystem developers. aws.amazon.com
GitHub Copilot documentation Official guide on AI Credits structure and plan limits. docs.github.com
