GitHub Copilot Moves to Usage-Based Billing on June 1, 2026

GitHub is rewiring how developers pay for Copilot. Starting June 1, every Copilot plan will move off premium request quotas and onto a metered system built around a new unit called GitHub AI Credits — one credit equals one U.S. cent of model spend, billed against your monthly allotment and any overage you buy.

The change touches every tier, from the $10/month Pro plan to Enterprise seats, and it lands at a moment when agentic coding has pushed token consumption far past what flat-rate pricing was designed for. The blunt summary from GitHub itself: the old model is no longer sustainable.

What's Actually Changing

The seat price is staying flat. Copilot Pro remains $10/month, Pro+ stays at $39/month, Business is $19/user/month, and Enterprise is $39/user/month. What's new is that each subscription now includes a monthly bucket of AI Credits — 1,000 for Pro, 3,900 for Pro+, 1,900 per user for Business, and 3,900 per user for Enterprise — that get consumed by chat, agents, and any premium model calls.

Credits are charged based on token usage (input, output, and cached) at each model's published rate. When your allotment runs out, paid plans can either set a spending cap and stop, or buy additional usage at the same $0.01-per-credit rate.

Critically, code completions and Next Edit Suggestions — the inline autocomplete most developers actually live in day-to-day — remain unlimited and don't draw credits. What does consume credits: Copilot Chat, the Copilot CLI, the Copilot coding agent, Copilot Spaces, GitHub Spark, and third-party coding agents wired in through the platform.

Why GitHub Says It Had To

The honest reason is that agent-mode and chat workloads burn tokens at rates that the previous "premium request" model couldn't price accurately. A single autonomous agent run can chew through more model spend in an afternoon than a heavy human-driven chat user racks up in a month.

Under request-based pricing, GitHub was effectively subsidizing the heaviest agentic users with the dollars from light users — and that math broke as agent adoption climbed. Moving to tokens lines up cost with consumption, but it also shifts the unpredictability of LLM spend onto developers and finance teams.

GitHub is launching a preview bill experience ahead of June 1 so admins can see projected costs before the cutover, available from the Billing Overview page on github.com. Power users and platform teams should turn it on this week.

Developer Reaction: Mostly Unhappy

Reaction from developers on the GitHub community discussion and across forums has been sharp. The most-cited complaint: a user who said the shift "could sober up an alcoholic from mere shock" and predicted that "you will get less, but pay the same price." A second common thread is unpredictability — request quotas were easy to reason about, while token spend depends on context size, model choice, and how chatty an agent decides to be.

Teams that lean on Pro+ and agent mode look set to pay noticeably more, and many of them haven't run the math yet. Several engineers in the threads said they're now reconsidering alternatives — kilo.ai with OpenRouter, Cursor, and Anthropic's Claude Code being the names that keep coming up.

There's a real risk of churn at the high end if the credit allotments don't comfortably cover typical agentic work. GitHub will need to publish concrete usage benchmarks fast — "how many agent runs does 3,900 credits actually buy?" is the question every engineering leader is now asking.

Why This Matters for the Industry

GitHub Copilot still anchors the AI coding market by seat count, so its pricing model effectively sets industry expectations. Anthropic's Claude Code, Cursor, Windsurf, and the Replit Agents have all been experimenting with credit-style billing for months; GitHub validating the model gives the whole sector cover to follow.

The real story is that agent workloads have officially outgrown SaaS-style pricing. Anyone running long-horizon AI agents — for code, sales, research, or ops — is going to face the same arithmetic: you can't charge a fixed price for an unbounded compute bill. Expect every coding-agent vendor to be on metered pricing by year-end.

For finance teams, the change creates a new line item that looks more like cloud spend than software spend. Variance month-to-month will be real, and the smartest engineering leaders will start treating Copilot the same way they treat AWS — with budgets, alerts, and per-team chargeback.

What to Do Before June 1

If you're an individual Pro user doing mostly autocomplete and occasional chat, you'll likely never touch the credit cap. If you're a team lead running agents, three things to do this week: enable the preview bill experience to model your actual spend, set per-user spending caps to avoid runaway runs, and benchmark a few common agent workflows in credits-consumed so you can plan budgets.

Admins on Business and Enterprise plans should also pull historical premium-request data and translate it into projected credit spend. If your power users are routinely going past their quotas now, the new pricing will hurt — and it's worth having that conversation before the bill lands in July.

Bottom Line

GitHub Copilot's move to usage-based billing is the single biggest pricing change in AI-coding tools to date, and it formally ends the flat-rate era for serious agent work. Seat prices aren't going up, but the bill probably will — especially for teams that have been running agents at scale on the old plans. The clock to model your spend runs out on June 1.