GitHub Copilot Switches to Token Billing, and Developers Balk

The flat monthly subscription, that comfortable fiction where you pay one price and use a tool as much as you like, is getting harder to sustain when the tool runs on someone else's expensive compute. GitHub learned this the hard way this month, and so did its customers.

On June 1, GitHub Copilot moved to usage-based billing. Instead of counting a fixed number of premium requests, every plan now comes with a monthly allowance of GitHub AI Credits, and your usage is measured in tokens: input, output, and cached, charged at each model's listed API rates. The full mechanics are laid out on the GitHub blog.

The bills that lit the fuse

For most people the change is quiet. Autocomplete, the feature that suggests the next line as you type, stays free and unlimited on every plan. If you mainly use Copilot for inline completions and the occasional chat question, the included credits should cover you, and the headline prices did not move. Pro is still $10 a month, Pro+ $39, Business $19 per user, Enterprise $39 per user, with each plan including a credit allowance equal to its price.

The pain is concentrated among heavy users of Copilot's agent features, where the model works autonomously across a codebase and burns tokens fast. Reports of bills climbing from $29 to $750, and in some cases from $50 to $3,000, spread across Reddit, X, and GitHub's own discussion threads within days, as Tech Times documented. Developers running heavy agentic sessions are projecting cost increases of ten to fifty times. GitHub also removed the fallback model that previously absorbed overflow usage, which sharpened the jump.

The mood was not subtle. TechCrunch summarized the developer reaction with a quote in its headline: "What a joke."

The real cost was always there

It is worth separating the sticker shock from the economics underneath. Agentic coding genuinely costs more to run than autocomplete, often by orders of magnitude, because the model reads, plans, edits, and re-reads across many files in a single session. Under flat pricing, light users were quietly subsidizing the heavy ones. Usage billing ends that arrangement, which is fairer in principle and more painful in practice for the people who had the better deal.

The timing is pointed. Only last week OpenAI recast its rival Codex as an office tool for every role, pushing agents deeper into daily work. As those agents do more, they cost more to run, and someone has to pay the meter. GitHub is the first of the big coding assistants to make that bill visible. It will not be the last.