GitHub will overhaul how it charges for its Copilot AI coding assistant starting June 1, 2026, replacing the existing 'premium request' model with a token-based credit system that ties costs directly to how much compute each user consumes — a shift the company says is necessary to sustain the platform as developers increasingly rely on multi-hour autonomous coding sessions.
GitHub announced Monday that all Copilot plans will transition to usage-based billing on June 1, 2026, introducing a new currency called GitHub AI Credits that tracks token consumption — including input, output, and cached tokens — rather than counting discrete premium request units.
The company stressed that base subscription prices are not changing. Copilot Pro remains $10 per month, Pro+ at $39, Business at $19 per user per month, and Enterprise at $39 per user per month. Under the new model, each paid tier will include a monthly allotment of AI Credits equal in dollar value to the subscription price — so a Pro subscriber gets $10 worth of credits per month — with the option to purchase additional usage beyond that allotment.
Why the change now
GitHub's engineering lead Mario Rodriguez pointed to a fundamental shift in how developers use Copilot. The product has evolved from a line-completion tool into what GitHub describes as an "agentic platform" capable of running long, multi-step coding sessions that iterate across entire repositories. Under the current flat-rate model, a quick chat query and a multi-hour autonomous session consume the same number of premium requests — a mismatch that GitHub says has made the economics increasingly difficult to sustain.
"Today, a quick chat question and a multi-hour autonomous coding session can cost the user the same amount," Rodriguez wrote. "GitHub has absorbed much of the escalating inference cost behind that usage, but the current premium request model is no longer sustainable."
What changes for users
Code completions and Next Edit suggestions will remain included in all plans and will not draw down AI Credits, meaning routine in-editor assistance is unaffected. However, one significant change will affect users who currently exhaust their monthly premium requests: the existing fallback experience — which allowed depleted users to continue working on a lower-cost model — will be discontinued. Instead, usage will be governed by available credits and admin-set budget controls.
Copilot's code review feature will also begin consuming GitHub Actions minutes in addition to AI Credits, billed at standard Actions per-minute rates.
To ease the transition, GitHub said it will launch a preview bill feature in early May, giving individual users and enterprise administrators visibility into projected costs before the June 1 cutover. The tool will be accessible through the Billing Overview page on github.com.
The company also noted that last week it introduced temporary usage restrictions on individual Copilot plans — including Free, Pro, Pro+, and Student tiers — and paused new self-serve Copilot Business sign-ups while it prepares its billing infrastructure. GitHub said those limits will be relaxed once the new system goes live.
The move reflects a broader industry trend: as AI model inference costs climb and usage patterns grow more complex, flat-rate subscription models that once attracted developers are giving way to consumption-based pricing more familiar from cloud infrastructure services.
Analysis
Why This Matters
- Developers who rely on heavy agentic workflows — running Copilot autonomously across large codebases — could see their effective costs increase significantly once the credit buffer is exhausted, even though headline subscription prices are unchanged.
- The removal of the fallback model experience means there is no longer a safety net for users who exceed their monthly allotment; work stops unless additional credits are purchased or an admin raises the budget cap.
- This signals a broader industrywide shift: AI coding assistant providers are moving away from the unlimited-use framing that drove early adoption and toward metered billing models more typical of cloud infrastructure.
Background
GitHub launched Copilot in 2021 as a flat-rate subscription built on OpenAI's Codex model, initially positioned as a simple autocomplete tool. As the underlying models grew more capable and expensive to run, GitHub introduced the concept of "premium requests" to differentiate standard from advanced model usage — an intermediate step that still did not fully capture the cost differential between lightweight queries and intensive agentic tasks.
The rise of agentic AI coding — where models autonomously plan, write, test, and iterate across entire projects with minimal human input — dramatically increased the compute footprint per user session. GitHub's parent company Microsoft has invested heavily in OpenAI and supplies the model infrastructure underpinning Copilot, but inference costs at scale have continued to pressure margins across the AI industry.
The June 1 transition is the most significant pricing architecture change since Copilot's commercial launch, and it arrives as competitors including Cursor, Windsurf, and JetBrains AI Assistant are also navigating how to price agentic workloads sustainably.
Key Perspectives
GitHub / Microsoft: The company frames the change as an alignment of pricing with actual value delivered, arguing that the current model unfairly subsidises the heaviest users at the expense of service reliability for everyone. The preview billing tool and unchanged base prices are positioned as evidence of a customer-friendly rollout.
Power users and enterprise developers: Developers who have been running agentic workflows under the assumption of a flat monthly cost may find their bills rising once credits are consumed. Enterprise admins gain new budget controls, but also new administrative overhead in managing credit allocation across teams.
Critics and skeptics: The removal of the fallback experience is a material downgrade for users who currently rely on it as a productivity floor. Some developers argue that the token-based model introduces unpredictability — it is harder to budget for a coding session that could consume variable amounts of tokens depending on model behaviour — and that GitHub's credit rates should be independently benchmarked against raw API pricing.
What to Watch
- The preview billing tool launching in early May will be the first real signal of how much costs will change for typical users; developer reaction on forums like Hacker News and Reddit will indicate whether the transition is broadly accepted or triggers a backlash.
- Whether GitHub relaxes its currently tightened individual plan limits as promised when June 1 billing goes live, or whether restrictions persist beyond that date.
- Competitor pricing responses: if Cursor, Windsurf, or others maintain flat-rate models, GitHub may face pressure on subscriber retention among cost-sensitive individual developers.