Microsoft ends unlimited GitHub Copilot usage as new token billing threatens enterprise developer budgets.

The era of subsidized AI software development is officially over. Starting June 1, 2026, Microsoft will strip away the unlimited buffet of GitHub Copilot and transition millions of users to usage-based token billing, exposing teams to premium model costs reaching $30 per million output tokens. What was once pitched as an all-you-can-eat coding assistant is evolving into a strict utility, forcing developers to fundamentally rethink how they deploy AI agents before runaway costs freeze their workflows mid-sprint.

For years, the tech industry has treated generative AI like a loss leader, subsidizing staggering compute bills to drive massive user adoption. That financial gravity is finally catching up. According to leaked internal documents, the week-over-week operational cost of running GitHub Copilot nearly doubled since January, prompting a sudden and urgent course correction.

To prepare its infrastructure for the pivot, Microsoft executed a sweeping freeze that rattled the developer community. In late April, GitHub completely paused new signups for its Pro, Pro+, and Student tiers, while simultaneously tightening rate limits for existing individual subscribers. Developers quickly took to forums and social media, arguing that the service degradation actively damaged their productivity.

"You will get less, but pay the same price," lamented users in a wave of critical feedback. The sentiment reflects a growing exhaustion with SaaS inflation, especially as developers realize that deep, agentic workflows will now require meticulous budget management.

Under the new paradigm, Copilot users will no longer operate on a frictionless, flat-rate model based on "premium requests." Instead, every prompt, codebase scan, and agent interaction will drain a fixed allotment of GitHub AI Credits. While base subscription prices remain untouched, $10 per month for Pro, $19 per user for Business, and $39 per user for Enterprise, the value derived from those plans will now be metered down to the exact input, output, and cached token.

"It better aligns pricing with actual usage, helps us maintain long-term service reliability, and reduces the need to gate heavy users," announced Mario Rodriguez, a product leader at GitHub, defending the controversial shift.

The financial calculus for engineering teams is about to become far more complicated. Copilot has evolved from a simple inline autocomplete tool into a heavy-duty platform where autonomous agents draft complex pull requests. These operations are incredibly token-intensive. A developer utilizing OpenAI’s high-end GPT-5.5 model will burn through credits rapidly, with costs pegged at $5 per million input tokens and a steep $30 per million output tokens. Lighter models, like GPT-5.4 nano, offer a budget-friendly alternative at just $0.20 for input and $1.25 for output.

This creates a terrifying scenario for ungoverned enterprise teams. Previously, when a user hit a limit, Copilot would silently fallback to a cheaper model without billing the customer extra. On June 1, that safety net vanishes. Once an organization depletes its pooled AI credits, they will either be charged published overage rates instantly, or developers will be completely blocked from utilizing AI assistance for the remainder of the billing cycle.

In preparation for the shock, GitHub is throwing a temporary financial bone to its highest-paying customers. Existing Copilot Business and Enterprise users will receive promotional pooled credits, $30 and $70, respectively, through the summer to help absorb the initial blow. Yet, the writing is on the wall. The industry-wide transition away from flat-rate billing marks the end of an artificially cheap era for AI developers.

Engineering managers now face a critical deadline to implement strict cost-control guardrails. The future of software development remains inextricably tied to AI, but the days of letting bots write code without checking the price tag are definitively over. If teams fail to adopt strict token accounting, their next cloud bill might just bankrupt their quarterly budget.