GitHub announced that all Copilot plans will transition to usage-based billing on June 1, 2026. The current system counts “premium requests” — a blunt unit that treats a one-line autocomplete and a two-hour autonomous coding session as roughly equivalent. The new system measures actual token consumption: input tokens, output tokens, and cached tokens, billed against a monthly allotment of “GitHub AI Credits.”
The headline numbers are unchanged. Copilot Pro stays at $10/month, Pro+ at $39/month. But the composition of what you get shifts. Code completions — the traditional inline suggestions — become explicitly free and unlimited, no longer drawing from any quota. What now consumes credits is anything agentic: Copilot in the coding agent, the code review bot, multi-step workspace tasks. The old fallback behavior, where Copilot would quietly downgrade to a cheaper model when you hit limits, is being removed.
The developer reaction has been predictably mixed. The most common complaint is that usage-based billing introduces unpredictability into what was previously a fixed monthly cost. GitHub’s community discussion thread opened with hundreds of responses, and the summary in outlets like Visual Studio Magazine — “you will get less, but pay the same price” — captures the sentiment accurately if uncharitably. GitHub’s counter is that a preview billing dashboard launches in May, giving users visibility into projected costs before any money moves.
GitHub’s underlying argument is correct even if the rollout is uncomfortable. The company says outright that a short chat question and a multi-hour autonomous coding session “can cost the same amount” under the current model, and that this is no longer sustainable as agentic workflows become the primary use case. That’s true. The flat-rate pricing made sense when Copilot was a smart autocomplete. It doesn’t scale to a system that can autonomously run tests, refactor codebases across files, and orchestrate multi-step tool calls for hours at a stretch. The old pricing was subsidizing the new capabilities, and that only works for so long.
What this represents, more broadly, is the first major pricing recalibration of the AI coding tool category. The flat monthly subscription was the hook that got millions of developers onto these platforms. Now that agentic features are the growth edge, the economics have to follow the compute. GitHub is the first major player to move explicitly to token-based billing, but it almost certainly won’t be the last — the underlying dynamics apply equally to Cursor, Windsurf, and anyone else offering autonomous coding sessions against a fixed monthly price.
Business and Enterprise customers get promotional bonus credits through August 2026, which is enough runway to evaluate actual usage patterns before the full model kicks in. For individual developers doing mostly traditional coding assistance, the practical impact will probably be small. For teams running Copilot agents on large repositories, it’s worth getting into the May preview tool early to see what the actual numbers look like.