What GitHub Copilot’s token-based billing means
GitHub Copilot’s new token-based billing is a usage-based pricing model where every AI interaction consumes metered AI Credits tied to underlying token costs across models and features, replacing flat subscriptions with direct, per-token charges that reflect real compute spending. Copilot originally launched as an in-editor autocomplete subscription, with a fixed monthly fee plus premium request units for advanced chat and longer “agentic” sessions. GitHub absorbed growing inference costs, but as autonomous coding sessions started spanning whole repositories, that structure became too expensive to maintain. Now each plan includes a monthly allowance of GitHub AI Credits, consumed on input, output, and cached tokens at published rates per model. Code completions and next edit suggestions remain included and do not spend credits, but most higher-end capabilities do. For many developers this means their bill now tracks how heavily they lean on Copilot’s smartest features instead of a simple flat rate.
How the AI Credits system and Copilot Max work
Under the new GitHub Copilot pricing, existing plan prices remain the same, but what you receive is now counted in AI Credits instead of premium request units. Individual Pro users paying USD 10 (approx. RM46) per month receive USD 15 (approx. RM69) in monthly credits, while Pro+ users paying USD 39 (approx. RM180) receive USD 70 (approx. RM324). Each plan includes base credits equal to the subscription price plus a flex allotment that GitHub says it will adjust as AI economics change. According to Microsoft VP Joe Binder, “The flex allotment is a variable part of your included usage; it is designed to adapt as the economics of AI evolve.” A new Copilot Max tier costs USD 100 (approx. RM460) per month and includes USD 200 (approx. RM920) in credits, aimed at developers running sustained, high-volume agentic workloads so they hit limits less often.
Why many developers are seeing cost increases
GitHub’s move to token-based billing exposes the real cost of longer, smarter AI sessions, and some users report their AI Credits disappearing faster than expected. In GitHub Community Discussions, one developer said their “12% of total AI credits burned” on minor updates while using Claude Sonnet, estimating around USD 0.35 (approx. RM1.61) per updated line. Another user shared a dashboard showing 3,705 credits remaining out of 7,000 after a single day’s work and complained that continuing the project might no longer be viable. Commenters argue the earlier flat subscriptions were effectively subsidized trials, with GitHub and Microsoft absorbing heavy losses as users ran large agentic jobs. As model sizes, context windows, and session lengths grow, that subsidy ends: requests that feel similar from a developer’s perspective may now consume far more tokens and credits, pushing many into higher effective monthly spend.
Budget controls for businesses and enterprises
For business and enterprise tiers, GitHub Copilot pricing still uses per-seat license fees, but AI Credits are pooled at the organization level instead of isolated per user. Each seat’s monthly credit allotment matches its per-seat price, and GitHub is temporarily adding promotional credits—USD 30 (approx. RM138) per Business user and USD 70 (approx. RM324) per Enterprise user—through August to soften the transition. Power users can draw heavily from the pool, while lighter users offset them, but this adds complexity. Organizations now configure four overlapping controls: a universal user-level budget, individual overrides, cost center budgets, and an enterprise-wide budget. User budgets always enforce a hard stop; the others apply when shared credits run out and metered overages begin. If total user budgets exceed the pool and the enterprise budget is too low, the “lowest remaining headroom wins” rule can block some developers earlier than their personal caps, so admins must revisit budgets whenever they raise individual limits.
Practical ways to control GitHub Copilot spending
To keep costs under control in a usage-based pricing model, developers need to treat AI Credits like any other finite resource. Start by tracking which activities consume credits: advanced chat, long-running agents, and code review now draw tokens, and code review also spends GitHub Actions minutes. Prefer standard completions and short prompts when possible, since they are included and cheaper in token terms. Watch credit dashboards daily at first to understand your burn rate, and set personal or team targets for monthly consumption. For organizations, set conservative user-level budgets and an enterprise budget that comfortably covers likely overages, then adjust as real usage data arrives. Encourage developers to batch complex tasks, reuse context, and avoid leaving agents running longer than needed. With careful monitoring and plan selection—including upgrading to Pro+, Max, or higher tiers only when justified—teams can use Copilot’s power without unpleasant billing surprises.
