By ByteBot
GitHub flipped the switch on June 1, 2026. Flat-rate subscriptions are out. Token-based “AI Credits” are in. Within hours, developers were reporting they had burned through half their monthly allotment in a single session. The GitHub Community announcement thread now sits at 958 downvotes and 24 upvotes. TechCrunch ran with the headline “What a Joke.” The Register documented mass cancellation threats. The backlash is real — and the math behind it is worth understanding.
What Actually Changed
GitHub replaced Premium Request Units (PRUs) with a new currency called GitHub AI Credits. One credit equals $0.01. Monthly plans now include a fixed credit allotment: Copilot Pro ($10/month) gets 1,500 credits; Copilot Pro+ ($39/month) gets 7,000; Copilot Max ($100/month) gets 20,000. Credits are calculated by token usage — input tokens, output tokens, and cached tokens — at per-model API rates. They do not roll over.
Code completions and Next Edit Suggestions remain unlimited on all paid plans. That is the original Copilot — the autocomplete product. It is still free to use as much as you want.
What is now metered: everything GitHub has been building toward. Chat interactions, agentic multi-step coding sessions, repository-wide iterations, Copilot code review. The features GitHub has spent two years promoting as the future of AI-assisted development are exactly what burns your credit budget fastest. Meanwhile, GitHub’s official announcement confirmed that budget controls and consumption tracking tools were not shipped at launch — so developers had no dashboard to understand their spending before the bills arrived.
The Numbers Behind the Outrage
Here is what developers actually reported in the first 48 hours after the GitHub Copilot billing change took effect:
- One developer consumed 822 credits — 54% of their Pro+ monthly quota — on a single request for a small project with a few screens.
- Another burned through 8% of their 7,000-credit Pro+ allocation in two hours of normal development work.
- Four agentic sessions in a single day cost 3,707 credits — 53% of a full month’s Pro+ allotment.
- Projected costs from community members: $29/month turning into ~$750/month; $50/month turning into ~$3,000/month.
The GitHub Community thread documents a 24x model price spread inside Copilot: MAI-Code-1-Flash costs $0.75 per million input tokens; GPT-5.5 costs $5 per million. A heavy iterative agentic session on a frontier model runs between $0.28 and $1.85. Run fifty of those sessions daily, and you are looking at roughly $2,000 a month per developer — on a plan that includes $39 in monthly credits.
The Agentic Trap
This is the part that stings. GitHub’s own justification for the billing change is that “Copilot is not the same product it was a year ago — it now powers far more complex, agentic workflows that consume far more compute.” That is accurate. However, it is also a precise description of the product GitHub deliberately built and marketed. They pushed developers toward agentic workflows. They built agent mode. They integrated Copilot code review. Then they restructured pricing on the assumption that developers would naturally moderate their usage — without providing any tools to track or predict consumption before the change went live.
Furthermore, the transition left out a critical detail: unused credits do not roll over. Every month, whatever you do not spend disappears. Combined with opaque per-session costs and no pre-launch budget visibility, developers essentially discovered what the new model cost them in real time, after the spending had already happened.
Where Developers Are Going
The complaints are not staying theoretical. Several migration paths are gaining serious traction:
- OpenRouter — runs in the same VS Code environment, credits roll over for up to a year, broader model selection. Some developers are using their Pro+ allotment until it runs out, then routing the rest of the month through OpenRouter.
- Cursor — flat subscription model, roughly 25% of the AI code editor market pre-billing change, now absorbing Copilot overflow signups.
- Claude Code — direct Anthropic subscription, ranked first in LogRocket’s February 2026 AI Dev Tool Power Rankings for complex reasoning tasks.
- Windsurf (now rebranded as Devin Desktop) — $15/month, positioned as a close Copilot substitute at lower cost.
- Cline and Continue.dev — open-source VS Code agents with bring-your-own-API-key models, effectively zero tool cost beyond direct API bills.
GitHub’s response to TechCrunch’s inquiry before publication: silence. The company has yet to make a public statement addressing the backlash directly.
The broader signal here is important. The era of unlimited AI coding subscriptions is ending — and GitHub is not the only company that will face this reckoning. Agentic workflows consume real compute at real cost, and the flat-rate model cannot survive at scale. The economic argument for usage-based billing is coherent. The execution — no rollover, no pre-launch tracking dashboard, no communication from the company as developers’ budgets evaporated — was not. If you are a Copilot Pro+ subscriber, check the models and pricing page now. Understand which models your workflows actually use. MAI-Code-1-Flash handles most everyday tasks competently at a fraction of GPT-5.5’s cost. That model selection decision is now a budget decision — and most developers did not know that until the bill arrived.
