GitHub Copilot's Usage-Based Billing Backlash

GitHub Copilot transitioned to token-based billing on June 1, forcing its 4.7 million paid subscribers to adapt to a fundamentally restructured pricing model. At midnight, the flat-rate premium-request model was replaced by GitHub AI Credits, under which each credit is priced at $0.01 and balances are deducted based on the number of tokens processed per interaction.

Opposition had been mounting since the initial announcement. The official FAQ thread on GitHub's community forum drew 435 comments and 904 downvotes against just 22 upvotes, marking one of the most lopsided responses in the platform's history. Among power users running autonomous coding agents, estimates circulated that monthly bills could increase tenfold to fiftyfold.

From Premium Requests to AI Credits: What Changed

The origin of these estimates lies in the new billing unit. In an April 27 blog post, GitHub announced the retirement of premium request units (PRUs) in favor of a monthly allotment of AI Credits for every plan. Under the new pricing, usage is calculated by applying the model's published API rates to input, output, and cached tokens.

While base subscription fees remain unchanged—Pro is priced at $10 per month with a $10 credit allowance, and Pro+ at $39 with a $39 allowance—these AI Credits now represent a hard cap rather than a soft ceiling. A Pro plan's $10 allowance equates to 1,000 credits, yet a single agentic session scanning a multi-file repository using a frontier model can consume 30 to 40 credits.

The transition also eliminates key safety nets. Previously, exhausting premium requests automatically downgraded users to a cheaper fallback model to ensure uninterrupted workflow, but this fallback has been discontinued. Once credits are depleted, premium features are suspended until the next billing cycle or a manual top-up. Furthermore, code reviews now simultaneously deplete both AI Credits and GitHub Actions minutes, fundamentally reshaping cost calculations. This shift to usage-based billing means heavier usage now translates directly into higher bills.

'Less for the Same Price': Developer Backlash

As the implications of the double-metering on code reviews became clear, developer sentiment turned increasingly critical. As Visual Studio Magazine reported, one user described the change as receiving 'less for the same price.' Others argued that the previous request-based system provided predictable pricing, whereas token-based metering varies continuously depending on the model, context window, and output length.

Projected figures intensified the frustration. TechCrunch reported that one Reddit user anticipated their monthly bill rising from $29 to nearly $750, while another expected a jump from $50 to approximately $3,000. Although GitHub and Microsoft have not verified these projections, and some observers attribute them to inefficient workflows, the community's lopsided response of 904 downvotes to 22 upvotes reflects the widespread discontent among Copilot users.

Online discussions were flooded with cancellations and posts declaring the platform's utility dead, alongside requests to restore access to Anthropic's Claude Opus and inquiries regarding whether unused credits roll over. Many developers questioned the value proposition of the service, noting that if billing is metered per token, invoking model APIs directly becomes a viable alternative. This skepticism mirrors the backlash that Google's Gemini limit overhaul faced, now redirected at GitHub Copilot.

Second Overhaul in a Year: GitHub Copilot Pricing at a Crossroads

The rapid spread of distrust stems from recent pricing volatility. Premium requests were only introduced in June 2025, meaning this is the second major billing overhaul in less than a year. The previous Copilot change capped Pro plans at 300 requests per month with a $0.04 fee per overage, resetting expectations just as users had begun to adapt.

The timing of the shift has also faced scrutiny. In his newsletter Where's Your Ed At, Ed Zitron cited internal Microsoft documents indicating that Copilot's operating costs had nearly doubled week-over-week since January. The data suggests the transition may be driven by immediate cost pressures rather than a long-term product strategy. Additionally, model multipliers for annual subscribers rose on June 1, with Claude Opus 4.7 consumption rates climbing from 7.5x to 27x.

Meanwhile, competing solutions are gaining traction. Coding environments like Cursor ($20 per month) and Windsurf ($15 per month) represent direct alternatives, while open-source tools such as Cline and Aider allow developers to use their own API keys to control inference costs. As AI compute costs increasingly rival engineering salaries, Microsoft CEO Satya Nadella has indicated that all of the company's per-user software offerings will transition toward usage-based models. This overhaul of Copilot appears to be the first step in that direction.