Developers who built workflows around GitHub Copilot now stare at bills that could multiply tenfold or more. The long-heralded AI coding assistant transitions to usage-based pricing on June 1. Flat monthly fees give way to charges tied directly to tokens consumed. And the reaction has been swift.
“What a joke,” one Redditor wrote after reviewing projected charges. The user had paid around $29 monthly. New estimates reached nearly $750. TechCrunch captured the outburst and similar complaints flooding Reddit and X in recent days. Another poster shared a screenshot showing costs jumping from roughly $50 to $3,000. “WOW, didn’t expect new pricing model to be this ridiculous,” the user added.
GitHub laid out the plan weeks earlier. Base subscription prices stay the same. Copilot Pro remains $10 per month. That now buys $10 worth of GitHub AI Credits. Pro+ at $39 includes $39 in credits. Business seats get $19 per user while Enterprise provides $39. Organizations receive extra promotional credits through August. Yet the meter runs on actual model usage. One AI credit equals one cent. Token counts for input, output and cached context determine the total. GitHub Blog spelled this out clearly on April 27.
Code completions and next-edit suggestions escape the new metering. They stay covered under the base plan without eating credits. Everything else — chat, agentic sessions, code review, CLI interactions — draws from the allotment. Once credits run dry, organizations pay for overages at published per-token rates. Individuals face the same structure. Unused credits vanish at month end. No rollover.
Model pricing varies sharply. Lighter options such as GPT-5 mini carry lower rates. Frontier models command premiums that reflect their compute demands. Official tables list costs per million tokens. Input for one versatile GPT variant runs $2 while output hits $8. Cached input drops to 50 cents. Similar spreads appear across Claude and other families. GitHub Docs publishes the full breakdown. A quick query might cost fractions of a cent. Hours-long agent sessions that spawn sub-agents and iterate across repositories burn through dollars fast.
That reality explains the fury. Many developers built habits during the subsidized era. Microsoft absorbed heavy inference costs to grow adoption. Agentic features encouraged long-running tasks. “Microsoft provided this billing method and they kept making it easier and easier to burn through massive numbers of tokens on single premium requests that could churn for hours or even days while spawning dozens or even hundreds of sub-agents,” one commenter argued. The company itself noted the product had changed. “Copilot is not the same product it was a year ago. It has evolved from an in-editor assistant into an agentic platform capable of running long, multi-step coding sessions,” the announcement stated.
Some voices push back against the outrage. They point to usage patterns. “The only way it gets crazy like that is if you are purely ‘vibe coding’ with a ton of bloated iterations,” one user observed. Proper tool application with lightweight models for simple questions and deliberate session management keeps spending reasonable. Even small teams can stay within budgets. Others marveled at prior losses. “Holy fuck how much money was copilot losing,” a Redditor asked bluntly.
The shift aligns with broader industry moves. Inference expenses have climbed. Subsidies that fueled rapid uptake now look unsustainable. Microsoft signaled the direction in earnings calls. Per-user products increasingly layer usage components. Copilot code review already migrated to an agentic setup that draws GitHub Actions minutes. Those minutes appear on the bill separately from AI credits starting next week.
Preparation tools rolled out in May. Users saw preview bills based on April activity. Admins gained dashboards showing consumption as both percentage of budget and dollar estimates. These surfaces appear inside editors, on GitHub.com and in management consoles. Budget caps can halt spend at organization, team or individual levels. The company urged teams to review historical data as a directional guide rather than precise predictor. Annual subscribers received different treatment. They keep the old premium-request model until renewal with adjusted multipliers that reduce effective allowances for premium models. After expiration they drop to free tier or switch to monthly usage-based plans.
Trials remain paused. GitHub cited abuse concerns and continues work on safeguards. Rate limits that frustrated users in recent months should ease once metering matches cost. The infrastructure for accurate per-token tracking required time to build. Short-term restrictions bought that window.
Enterprise customers gain pooled credits and higher promotional allotments. A Business organization with 10 seats receives $190 base plus $300 promotional during transition months. That buffer softens the landing for teams still calibrating agentic habits. Individual developers lack similar cushions. Pro and Pro+ users get exactly the credit value of their subscription. Heavy experimenters or those relying on expensive models will see overages quickly.
Conversations on X this week echoed the TechCrunch coverage. Posts highlighted screenshots, questioned Microsoft’s timing and speculated on alternatives like local models or rival services. One thread tied the change to Microsoft’s larger capital expenditures on data centers. Another noted Nadella’s comments about transforming per-user businesses into hybrid per-user and usage models across productivity, coding and security products.
Developers now face concrete choices. Monitor usage surfaces religiously. Favor lighter models for routine tasks. Break complex agent jobs into smaller steps. Set strict budgets. Some will cancel. Others will absorb the expense as cost of accelerated output. Enterprises may negotiate custom deals or optimize model selection at scale. The era of essentially unlimited agentic exploration under a flat fee has closed.
GitHub insists the move supports long-term reliability. Predictable economics reduce pressure to throttle power users. Teams gain transparency into true consumption. Whether that transparency comforts those staring at triple-digit jumps remains an open question. The complaints captured today suggest many feel blindsided despite preview tools and advance notice.
Microsoft has not commented publicly on the specific user examples circulating this week. The company’s blog post focused on evolution of the product and sustainability. Documentation walks through every calculation. Yet the human reaction — shock, anger, resignation — plays out in public forums less than 48 hours before the switch flips. Power users built workflows that the old model quietly subsidized. The bill has arrived.