GitHub Copilot switches to tokens

GitHub is changing what “using Copilot” means. Starting June 1, 2026, Copilot stops counting premium requests and starts charging by actual model usage — measured in tokens, then converted into GitHub AI Credits. That sounds like a billing tweak. It isn’t. It’s GitHub admitting that Copilot is no longer mostly autocomplete — it’s becoming a heavier, more agentic product, and the old pricing model was breaking under that load. ### What actually changes on June 1? Every Copilot plan moves to usage-based billing. Instead of a monthly bucket of premium requests, plans get a monthly allowance of GitHub AI Credits. Those credits are consumed from token usage — input tokens you send, output tokens the model generates, and cached tokens from reused context. Paid plans can buy more usage once they burn through the included allowance. ### What’s an AI Credit in plain English? It’s GitHub’s billing wrapper around token costs. GitHub says 1 AI Credit equals $0.01 USD. So the real meter is still tokens, but GitHub surfaces the spend in credits so different models and features can roll up into one usage number. Basically, Copilot is moving closer to cloud-style metering — use more compute, spend more money. ### Why ditch premium requests now? Because premium requests stopped matching reality. GitHub said agentic workflows have fundamentally changed Copilot’s compute demands, and in late April it paused some new sign-ups, tightened usage limits, and adjusted model availability for individuals to protect service quality. The old request-based model and task. ### Are basic completions still free? Mostly, yes — but the details matter. GitHub’s docs say all plans, including Copilot Free, come with some monthly AI Credit allowance. The practical shift is that “free” now means “free up to an included usage cap,” not unlimited lightweight assistance. If you stay inside the allowance, it's fast. ### Who should worry most? Teams with heavy Copilot usage, especially where developers use chat, agents, code review, and larger frontier models all day. Those workflows chew through more tokens than classic inline completion. GitHub is also adding a second meter for Copilot code review on June 1 — AI Credits for the m some orgs could see costs rise from two directions at once. ### Is there any weird transition catch? Yes — annual individual subscribers aren’t all flipping the same way immediately. GitHub’s docs note that some existing Copilot Pro and Pro+ annual plans can stay on request-based billing for a period, but model multipliers change after June 1. So there isn’t one universal migration path. The headline is simple, but the edge cases are not. ### What should teams do before June 1? Turn on usage visibility and budget controls now. GitHub’s organization billing docs point admins to preparation guidance, usage tracking, and spend estimation by model. The smart move is to find the expensive behaviors before the pricing switch, not after the invoice. Think of it like moving from an all-you-can-eat buffet — every extra spoonful has a price. ### Bottom line This is GitHub bringing Copilot back in line with the economics of modern AI. For light users, the change may barely register. For power users and startups with lots of agent-driven coding, June 1 is the date when “Copilot usage” turns from a fuzzy subscription perk into a line item you actually have to manage.