GitHub meters Copilot with credits
- GitHub said all Copilot plans switch to usage-based billing on June 1, 2026, replacing premium requests with GitHub AI Credits for individuals, teams, and enterprises. - One AI Credit equals $0.01, and usage now tracks input, output, and cached tokens by model—while annual Pro users face new multipliers if they stay put. - This turns Copilot from a flat-ish seat cost into a metered compute budget—especially for agentic workflows and shared enterprise spending.
GitHub Copilot billing is changing from a simple-ish request counter into a meter. Starting June 1, 2026, Copilot usage will consume GitHub AI Credits across every plan, from free individual accounts to enterprise rollouts. That sounds like a pricing footnote, but it’s really a product shift. GitHub is saying the old model no longer matches how people use Copilot now that chats are longer, models are mixed, and agents can take multiple steps to finish one job. ### What is GitHub actually changing? Today’s Copilot has been billed partly through “premium requests” — a rough unit that treated one advanced interaction as one billable event, with model multipliers layered on top. GitHub is replacing that with token-based usage. Every interaction now counts the tokens you send in, the tokens the model sends back, and cached tokens from reused context, then converts that total into AI Credits. ### What’s an AI Credit? Basically, it’s GitHub’s internal billing chip. GitHub’s docs peg 1 AI Credit at $0.01 USD. The important part is not the penny number — it’s that the charge now maps to actual model consumption instead of a blunt request bucket. Bigger prompts, longer outputs, and heavier models cost more. Smaller interactions cost less. Why did the old system stop fitting? Because Copilot stopped being just autocomplete. Agentic workflows changed the math. A coding agent can inspect files, plan steps, call tools, revise code, and explain itself — all inside what feels like one task to the user. Under request-based billing, that kind of work was hard to price cleanly. GitHub said the move is meant to align pricing with actual usage and make the business sustainable and predictable. ### Who feels this most? Organizations do. GitHub’s docs say Copilot Business and Enterprise plans include per-user AI Credit allowances, but those allowances are pooled at the billing-entity level. In plain English, one team’s heavy usage can draw from the same shared budget as everyone else. That makes Copilot look less like a fixed software seat and more like cloud spend — something finance and platform teams will watch month by month. ### What about individual annual subscribers? There’s a carveout. Copilot Pro and Pro+ users on existing annual plans can stay on request-based billing after June 1, 2026, but GitHub says the model multipliers will change if they do. If they switch to monthly usage-based billing instead, those multiplier changes do not apply. GitHub also offered prorated refunds for some users making that switch during the transition window. ### Why is GitHub showing previews now? Because this is the kind of change that can surprise people. GitHub has started surfacing preparation tools in early May, including bill previews that compare current spend with estimated AI Credit charges based on April 2026 usage. That gives admins a dry run before the June 1 switchover. It’s a classic cloud-pricing move — show the future bill before the meter goes live. ### What else gets metered now? Copilot code review is a good example. Starting June 1, 2026, GitHub says each code review will consume AI Credits under the new billing model, and the feature can also consume GitHub Actions minutes because the review workflow runs on Actions infrastructure. So one “Copilot feature” can now touch more than one budget line. ### So what’s the real takeaway? The headline is billing, but the deeper story is control. GitHub is making Copilot legible as compute. That means teams can finally ask a harder question than “do we have licenses?” They can ask whether a given workflow is worth its token bill. For hobby users, that’s annoying but manageable. For companies building AI into daily engineering work, that’s the beginning of cost-per-outcome thinking.