GitHub Copilot Ditches Flat-Rate Pricing: What Usage-Based Billing Means for Developers

On June 1, 2026, GitHub retired the pricing model that had defined its AI coding assistant since launch. Premium Request Units — the discrete-credit system developers had used to understand their Copilot usage for years — are gone. In their place: GitHub AI Credits, a token-metered system that bills based on actual model consumption rather than a simple count of requests. The subscription sticker prices did not move — Copilot Pro remains $10 a month and Pro+ remains $39 — but the mechanics underneath them changed completely, and for a meaningful subset of developers, the effective cost of using Copilot the way they always have just went up substantially.

How the Old System Worked, and Why It Broke

Under the previous model, every interaction with a premium AI model in Copilot counted as one “premium request,” with a multiplier applied depending on which model was used. It was a simple mental model for developers: a fixed number of premium requests included each month, consumed one at a time. The problem, according to GitHub’s own explanation for the change, is that Copilot in 2026 is simply not the same product it was when that system was designed. Agentic workflows — where Copilot autonomously plans, edits across multiple files, and iterates over long sessions — consume dramatically more compute per interaction than a single autocomplete suggestion or one-shot chat question ever did. GitHub has said the shift is intended to align pricing with actual usage and cost, describing it as necessary for a more sustainable and reliable product experience going forward.

What Changed, Concretely

The rollout followed a clearly telegraphed timeline. GitHub first flagged the coming change on April 17, 2026, in a community post about upcoming changes to Copilot Individual plans, followed by a formal company blog post on April 27 laying out the full usage-based billing plan. In the interim, GitHub also temporarily paused new self-serve signups for paid Individual plans and rolled out short-term reliability adjustments as it prepared its infrastructure for the transition. A preview billing experience went live in early May, letting existing users see projected costs under the new model before it actually took effect on June 1.

Under the new system:

Copilot Pro ($10/month) includes a monthly allotment of AI Credits aligned to that price — roughly $10 to $15 in credits, depending on the source consulted.
Copilot Pro+ ($39/month) includes a larger allotment, cited variously as $39 to $70 in credits.
Each AI Credit is worth exactly $0.01, and consumption is calculated from real input, output, and cached token usage at the published rate for whichever model is in use.
Once a user’s monthly credit allotment is exhausted, additional premium requests cost roughly $0.04 each, or access is paused until the next billing cycle, depending on plan settings.
Inline code completions and next-edit suggestions remain unlimited and entirely outside the new credit system — a detail GitHub has emphasized, since it means developers whose usage is overwhelmingly simple autocomplete are largely unaffected by the change.

Annual-plan subscribers were handled slightly differently from monthly subscribers during the transition. Monthly Pro and Pro+ users were automatically migrated to usage-based billing on June 1. Annual subscribers, by contrast, were allowed to remain on their existing premium-request-based pricing until their plan term expires, though GitHub did apply increased model multipliers to annual plans starting the same day, meaning even customers who kept the old billing structure saw its underlying math shift.

The Developer Reaction Was Immediate — and Largely Negative

The response from GitHub’s developer community was swift. Within hours of the change going live, complaints and criticism were trending across GitHub’s own community discussion boards, with one heavily upvoted comment describing the new pricing as a direct hit against individual users specifically. The core grievance is not really about the headline subscription price — that stayed flat — but about predictability and included value. Developers leaning heavily on Agent Mode, multi-file editing sessions, and automated code review are finding that their monthly credit allotment simply does not stretch as far as their old premium-request allotment did, even though they are paying the exact same monthly fee.

One trade publication summarized the developer sentiment bluntly, capturing widespread frustration that the change means paying the same price for less usable capacity in real-world agentic workflows.

GitHub’s own community FAQ thread, updated repeatedly through May and June, has fielded a wide range of concerns from users — about reduced included value compared to the old system, about the unpredictability of token-based costs versus a simple request count, about which specific models remain accessible under the new structure, about whether unused credits roll over between billing cycles, and about whether Copilot’s overall value proposition still holds up against directly using foundation-model APIs or competing coding assistants.

Who Is Actually Affected — and Who Isn’t

The honest answer, according to multiple analyses published in the weeks following the transition, is that the impact is genuinely bimodal. Most individual developers and teams, whose usage skews toward lighter chat interactions and standard autocomplete, are not going to see a materially different bill under the new system. But a smaller subset — heavy Agent Mode users, teams running frequent automated code review, and anyone leaning on the most expensive frontier models for extended sessions — are seeing meaningfully higher effective costs, and in many cases do not yet realize it until their first usage-based bill arrives.

One detail worth understanding clearly: a single request under the new system is not a fixed unit. Depending on the model and the size of the request — how much context is loaded, how long the output runs — a single interaction can consume anywhere from roughly one credit to as many as two thousand, according to one billing-focused analysis tracking the rollout. That range is precisely why the shift feels so different from the old premium-request system, where every request of a given model type cost the same predictable amount regardless of its actual size.

The Business and Enterprise Picture

For organizational customers, GitHub has largely directed Business and Enterprise buyers toward sales conversations for current per-seat credit allotments rather than publishing a single universal figure, reflecting the more customized nature of large-scale Copilot deployments. Analysts tracking the transition note that the shift changes the fundamental unit organizations need to plan around — from a straightforward seat count to a usage-shape calculation that depends heavily on how much of an organization’s engineering work involves agentic, multi-step Copilot sessions versus lighter completion-style usage.

Practical Guidance for Developers and Teams

For anyone trying to figure out whether this change actually costs them more, a few concrete steps are worth taking:

Use the preview billing dashboard. GitHub’s Billing Overview page now shows projected usage-based costs, giving a real, personalized read on how the new system affects your specific usage pattern rather than relying on general commentary.
Separate autocomplete-heavy usage from agentic usage. Since inline completions remain unlimited and free of the credit system, developers whose workflow is mostly quick suggestions have far less to worry about than those running extended Agent Mode sessions.
Watch your model selection. Because cost is now tied to which specific model handles a given request, teams that default to the most capable — and most expensive — frontier model for every task will see costs climb faster than teams that match model choice to task complexity.
Reassess the build-versus-buy calculation. With Copilot’s value proposition now more directly comparable to metered API pricing from foundation-model providers, some teams are recalculating whether Copilot’s bundled convenience still beats assembling their own tooling directly against model APIs.

What Comes Next

GitHub has signaled it will loosen usage limits further now that usage-based billing is fully in effect, suggesting the initial rollout period involved some deliberately conservative caps while infrastructure adjusted to the new load pattern. Whether that translates into meaningfully more headroom for heavy users, or simply a modest easing at the margins, is the detail developer communities are watching most closely heading into the second half of 2026. For now, the clearest takeaway is that Copilot’s headline price is no longer the whole story — the real cost now depends entirely on how, and how intensively, each individual developer actually uses it.

Frequently Asked Questions

Did GitHub Copilot get more expensive? Not on paper — the listed subscription prices for Pro and Pro+ are unchanged. In practice, whether it costs more depends entirely on usage pattern. Autocomplete-heavy users are largely unaffected; heavy Agent Mode and code-review users are generally seeing less effective capacity for the same monthly fee.

What happens if I run out of AI Credits mid-month? Depending on your plan configuration, you either pay approximately $0.04 per additional premium request beyond your allotment, or your premium access pauses until the next billing cycle begins.

Are inline completions billed under the new system? No. GitHub has kept inline code completions and next-edit suggestions unlimited and entirely separate from the AI Credit system, regardless of plan tier.

Is this change permanent, or could GitHub adjust it further? GitHub has already indicated it plans to loosen usage limits now that the new system is fully live, suggesting the initial credit allotments and caps may be adjusted upward as the company gets more real-world usage data. Developers should expect further tuning rather than treating the June 1 configuration as final.

Tedony will continue following how GitHub adjusts Copilot’s usage limits and credit allotments as the new billing model matures.

GitHub Copilot Ditches Flat-Rate Pricing: What Usage-Based Billing Means for Developers