GitHub Copilot’s Token Billing: What the $180 Bill Means

GitHub quietly replaced flat-rate Copilot pricing with token-metered billing called GitHub AI Credits on June 1, causing one developer on the $39/month Pro+ plan to receive a $180 bill and others to project cost jumps of 10x to 50x. The change, which applies to chat, agent mode, code review, and other features but not inline code completions, means agentic workflows can consume an entire monthly credit budget in a single session, with overage charges of $0.01 per credit. GitHub and Microsoft made the switch to cover rising inference costs as agentic AI usage scales, a move expected to be followed by other AI tool providers.

On June 1, GitHub quietly ended flat-rate Copilot pricing. One developer on the $39/month Pro+ plan received a $180 bill on day one. Another burned through 8% of their monthly AI Credit allowance in two hours and estimated they would hit zero before the week was out. The math was buried in the changelog. The sticker shock was not. What Actually Changed GitHub replaced the old “Premium Request Units” model with token-metered billing called GitHub AI Credits . One credit equals one cent. Usage is calculated from token consumption — input, output, and cached tokens — multiplied by the model’s per-token rate. The switch took effect June 1 across all plans. The one thing that did not change: inline code completions. Those remain free and do not consume AI Credits. Everything else — chat, agent mode, code review, Copilot CLI, the cloud agent — now draws from your monthly credit pool. Here is what each plan includes: | Plan | Price | Credits Included | |---|---|---| | Copilot Pro | $10/month | 1,500 credits | | Copilot Pro+ | $39/month | 7,000 credits | | Copilot Business | $19/user/month | 1,900 credits/user | | Copilot Enterprise | $39/user/month | 3,900 credits/user | One trap that has not been widely communicated: model choice now determines your burn rate. GitHub’s cheapest model MAI-Code-1-Flash and its most expensive GPT-5.5, at $5 input / $30 output per million tokens have a 24x price gap . If you are defaulting to a frontier model for every chat, you are paying frontier prices per message. The Math Nobody Told You About The problem surfaces in agentic workflows. A single agent mode session working across a medium-sized codebase costs roughly $30 to $40. On the Pro+ plan, your monthly credit budget is $39. That means one productive agentic session can consume your entire month’s included allotment — and the month just started. For Business plan users, the situation is worse. At $19/user/month, you get 1,900 credits per seat. Two agentic sessions a week runs somewhere between 25,000 and 30,000 credits monthly. The overage alone — billed at $0.01 per credit — works out to $230 to $280 on top of your subscription. One developer reported spending 1,180 credits on a single Claude 4.8 session and described the results as “mediocre.” That session cost nearly two-thirds of a Business plan’s monthly budget. Reports on Reddit and Hacker News show projected cost jumps of 10x to 50x for power users https://www.theregister.com/ai-and-ml/2026/06/02/github-copilot-users-threaten-exit-as-metered-billing-kicks-in/5249826 . Some are projecting monthly bills going from $29 to $750. One team lead shared a scenario where their team’s agentic usage would push bills past $3,000 per month. Why GitHub Did This The flat-rate model was always a subsidy. GitHub and Microsoft were absorbing inference costs to drive adoption during a land-grab phase. Agentic AI workflows are genuinely expensive to run — multiple model calls, long context windows, tool invocations. Charging $39/month while users run unbounded agentic sessions made no business sense at scale. This is not unique to GitHub. It is what happens when AI tools exit the promotional phase and enter managed infrastructure. Microsoft needs to cover inference costs, and the margins on subsidized flat-rate plans were never there. Expect other tools to follow. Copilot went first — or at least got caught first. What To Do Right Now A few practical steps worth taking before your next billing cycle: Set a spending limit. GitHub launched budget controls https://docs.github.com/en/copilot/tutorials/budgets/getting-started-with-budget-controls at the user, cost center, and enterprise level. Use them. Without a cap, overages bill automatically. Check which model you are using. Agent mode defaults to a frontier model. Switching to MAI-Code-1-Flash for non-critical tasks can dramatically reduce burn rate. Reserve agent mode for substantive work. Chat is cheaper. Completions are free. Use them for quick lookups and let agent mode handle the tasks that actually warrant the cost. Check your usage dashboard now. Do not wait until the end of the month to discover how fast credits are moving. Alternatives Worth Considering Cursor and Claude Code both sit at $20/month with flat-rate agentic access — roughly half the cost of Copilot Pro+ before any overages. For teams heavy on agentic workflows, GitHub’s own announcement https://github.blog/news-insights/company-news/github-copilot-is-moving-to-usage-based-billing/ concedes the new billing “will be higher for users who use agent mode frequently.” If that is your team, the economics have shifted. For privacy-first or budget-constrained teams, OpenCode MIT licensed, self-hosted, bring-your-own-API-keys has been gaining stars rapidly — it bypasses the Copilot markup entirely and lets you pay model providers directly. GitHub Copilot still has genuine advantages: mature enterprise controls, SSO, audit logs, and deep native integration with GitHub workflows — PRs, code review, the Copilot App. Those matter more to some teams than monthly cost. But the math has changed, and pretending otherwise is expensive. Check your model pricing page https://docs.github.com/en/copilot/reference/copilot-billing/models-and-pricing before your next sprint kicks off.