GitHub is finally admitting what everyone suspected: unlimited AI coding for a flat fee was a mirage. Starting June 1, Copilot users will be billed by actual token consumption, not by vague “requests” or “premium requests.” The Microsoft owned company claims it can no longer absorb the “escalating inference cost” caused by power users running agentic assistants like Openclaw, which can burn through tokens in hours. This is the end of the all-you-can-eat AI buffet.
The Fine Print: Token Taxonomies and GPT Price Hikes
Under the new scheme, you get a monthly allotment of “AI Credits” tied to your subscription level. Go over that, and you pay API rates per token. Those rates are brutal: GPT-5.5 costs $30 per million output tokens, while the cheaper GPT-5.4 Mini runs $4.50. Code completions and Next Edit suggestions remain free, but Code Reviews now consume GitHub Actions minutes. GitHub has released a “preview bill” tool to show users how badly they will be gouged, but it is little comfort for developers who relied on flat rate pricing to hack endlessly. Expect similar pain from Anthropic, which recently began charging enterprise Claude customers the full compute cost. The era of subsidized AI is over.
The Real Story: Profit Over Promise
This is a blatant cash grab dressed up as sustainability. Internal documents leaked to critic Ed Zitron show Copilot’s week-over-week costs nearly doubling since January, driven by agentic workflows. GitHub is not trying to be fair. It is trying to stop its most valuable users from running up bills that undermine Microsoft’s cloud margins. The same dynamic is playing out across the industry. Anthropic briefly removed Claude Code from its Pro plan and now throttles usage during peak hours. These companies raised billions on promises of AI abundance. Now they are rationing the candy. The message is clear: you use AI, you pay. No more free coding. No more subsidies. Just a meter running behind every prompt.
Source: Arstechnica
