GitHub’s Metered Copilot: Predictability Dies

When the Meter Runs, Trust Runs Out

GitHub officially moved Copilot to usage-based billing on June 1, 2026, replacing flat subscriptions with AI Credits consumed per token. The backlash was immediate and brutal: developers reported burning through a month’s credits in hours, with one Pro+ user ($39/month) watching 8% of their 7,000-credit quota vanish in two hours of work. Another developer made a single refactoring request and burned $6. The core issue isn’t the price table—it’s that predictable consumption just died.

The new model works like this: code completions remain unlimited, but chat and agentic features now drain monthly AI Credit allotments based on token usage (input, output, cached). Pro costs $10/month with $15 in credits; Pro+ is $39 with $70; the new Max tier is $100 with $200. Business and Enterprise get pooled credits—1,900 and 3,900 per user respectively, with promotional bumps to 3,000 and 7,000 through September 1. One AI Credit equals one cent.

What GitHub framed as “pay for what you use” feels more like “guess wrong and pay a surprise tax.” The problem is that token consumption is invisible and variable. A simple “refactor this function” might cost pennies or dollars depending on context window size, model selection, and how Copilot interprets the request. Developers don’t think in tokens; they think in tasks. When a two-hour coding session threatens to exhaust a monthly budget, the tool stops being an assistant and becomes a cognitive tax—you start second-guessing every question, every multi-file edit, every exploratory prompt.

The timing is also suspect. GitHub spent years positioning Copilot as a flat-rate productivity multiplier, building user habits around “just ask.” Now that developers are locked in—workflows built, muscle memory trained—the pricing model flips to consumption-based, and suddenly that frictionless experience has friction everywhere. It’s the classic SaaS bait-and-switch: hook users on predictability, then extract rent on unpredictability once they can’t easily leave.

And the alternatives? Cursor, Codeium, and a dozen other AI code assistants still offer flat-rate or freemium models. GitHub is betting that integration lock-in (it’s built into VS Code, deeply tied to GitHub repos) will keep users paying even as costs balloon. But developer loyalty is thin when the value prop reverses. If I’m now optimizing my questions to save credits instead of optimizing my code to solve problems, the tool has failed its job.

The real tell is in GitHub’s response. When pressed on complaints, a spokesperson pointed to “spending limits, usage dashboards, and model selection” as cost management tools. Translation: you now need a budget dashboard to use a code completion tool. That’s not productivity—it’s overhead.

Here’s what GitHub should have done: cap the metered tier. Charge $10 for the first $15 of usage, then hard-stop at a ceiling (say, $60 total) where unlimited kicks in. That preserves predictability for power users while letting light users pay less. Instead, we got uncapped metering with promotional windows and tier upsells—a model optimized for revenue extraction, not developer trust.

The larger lesson: metered billing works when usage is discretionary (run this one-off batch job), but fails when usage is habitual and integrated into thought process. You can meter API calls; you can’t meter the act of thinking out loud to an AI pair programmer without turning coding into cost accounting.

FAQ

Q: Can I still use GitHub Copilot on a flat rate?
A: No. All monthly plans transitioned to AI Credits on June 1, 2026. Annual Pro/Pro+ users stay on the old request-based model until renewal, then switch to monthly metered or downgrade to the free tier.

Q: What happens when I run out of credits mid-month?
A: Paid plans can purchase additional credits at $61 per credit. Free tier users lose access to chat/agents until the next billing cycle, though code completions remain unlimited across all tiers.

Q: Is there any way to predict my monthly Copilot cost under the new system?
A: Not reliably. Token consumption varies by task complexity, context size, and model. GitHub provides usage dashboards, but they’re reactive—you only know you overspent after the fact. The only true cost ceiling is upgrading to Max ($62/month with $63 credits) and hoping that’s enough.