← Back to feed

GitHub officially announced Monday that every Copilot plan will switch to usage-based billing on June 1, 2026. The flat-rate subscription model that charged for seat access and counted “premium requests” is being replaced by GitHub AI Credits consumed at per-token rates.

How the New Model Works

Every Copilot plan — including the free tier — will receive a monthly allotment of GitHub AI Credits. Paid plans can purchase additional credits when the allotment runs out.

Credit consumption is calculated on token usage: input tokens, output tokens, and cached tokens, each billed at the listed API rate for the specific model used. Switching to a more capable model costs more per task. Using cached context costs less. Volume shapes the bill.

GitHub will launch a preview billing experience in early May, giving users and administrators a projected cost view before the June 1 transition hits.

Why This Happened

GitHub stated the change directly: Copilot is not the same product it was a year ago. The shift from in-editor autocomplete to multi-step agentic coding — where an agent runs continuous sessions, invokes tools, reads large codebases, and generates hundreds of thousands of tokens per task — made flat-rate pricing structurally unsustainable.

A developer using Copilot for autocomplete might generate a few thousand output tokens per session. A developer running Copilot coding agents on a complex refactor generates that in a single round-trip. A flat monthly fee cannot price those use cases in the same bucket without either undercharging heavy users to the point of loss or overcharging light users to the point of churn.

The Next Web characterized the situation as “agentic AI breaking the economics of flat-rate developer subscriptions.” That is the correct framing. Agentic use patterns create an unbounded consumption tail that subscriptions cannot absorb.

Context: The Preceding Moves

GitHub had already moved to pause new sign-ups for Pro, Pro+, and Student plans and tighten usage caps — a defensive step while the pricing transition was being prepared. Monday’s announcement formalizes what the cap changes were signaling: the flat-rate era is ending, not just constrained.

The billing change aligns Copilot’s model with the broader market direction. OpenAI’s Codex moved to a metered per-task pricing structure. Anthropic’s Claude Managed Agents charges $0.08 per agent-hour. Token-level billing — not seat licensing — is becoming the standard pricing unit for AI developer tools as agent session lengths detach from human-scale interaction patterns.

What It Means for Enterprise

For enterprises running Copilot across large engineering teams, the June 1 shift introduces a variable cost line where there was previously a fixed one. The preview dashboard is the critical tool: teams will need to model their actual agent usage patterns before June 1 to avoid budget surprises.

The credit allotments per plan tier have not been published as of this writing. GitHub said more detail will accompany the May preview launch.

The model-level pricing detail — different credit rates for different Copilot models — also suggests GitHub is laying groundwork for differentiating which models are available at which price points, extending the pattern set earlier when it restricted access to Opus 4.7 to paid Pro+ tiers.

Flat-rate AI subscriptions served autocomplete. Token-based billing serves agents. GitHub has picked its side.