GitHub Copilot is about to undergo its most significant pricing overhaul since launch. Starting June 1, 2026, Microsoft will shift the developer tool from a flat monthly subscription to a per-token billing system. This aligns Copilot with how companies like OpenAI and Anthropic already charge enterprise customers for large language model access. The change marks a departure from the simpler, request-based model that developers have grown accustomed to.

Under the current system, users get a set number of premium requests per month. A complex, multi-hour coding task counts as one request. A trivial query asking for a function name also consumes one. This all-you-can-eat approach for predefined requests is about to vanish, replaced by something far more granular and potentially costly for heavy users.

How Token Billing Works for Developers

Tokens are the new currency in AI-assisted coding. One token represents roughly three-quarters of a word in natural language. In code, that translates to about 12,000 to 13,000 tokens for every 10,000 logical units, such as variable names, expressions, or function calls. Both inputs, like your prompt or code snippet, and outputs from Copilot will count toward your monthly allowance.

GitHub will still charge the same monthly fee for its tiers, but instead of a fixed number of queries, users will receive AI Credits. A Copilot Pro subscriber paying $10 per month gets 1,000 credits, each currently worth one cent. How many tokens each credit buys depends on several factors: the specific model used, the ratio of input to output, cache size for context, and the complexity of the feature requested.

The Practical Impact on Your Workflow

If you mostly ask simple questions or request short code completions, you might never need to purchase extra credits. But consider a developer debugging a sprawling legacy codebase with tens of thousands of lines. A single multi-agent query analyzing that code could burn through credits surprisingly fast. Queries directed at frontier models, the most advanced and power-hungry ones, will cost more than those using lighter alternatives.

There is some good news. Code completions, similar to your phone’s autocomplete feature, and Next Edit suggestions will remain free. Microsoft is trying to soften the blow for everyday tasks while charging for the heavy lifting. Still, the psychological shift is real. Developers accustomed to thinking in terms of requests must now track token consumption per query, a figure previously hidden behind the monthly subscription abstraction.

Why Microsoft Is Making This Move Now

Unlike OpenAI or Anthropic, Microsoft is a profitable conglomerate. It has historically subsidized Copilot with revenue from Azure, Office, and Windows. Until now, users could effectively spend between three and eight times the token value their subscription covered without penalty. That grace period ends on June 1.

From a business perspective, this pricing change makes sense. AI inference costs scale with usage, not subscriptions. By moving to per-token billing, Microsoft aligns revenue with actual compute consumption. It also nudges the industry toward a standard that large enterprises already expect. Yet for individual developers and small teams, the switch discourages experimentation. Exploring new features or testing edge cases now carries a direct cost, which may slow adoption among those still evaluating the tool.

Industry-Wide Shift and Real-World Consequences

GitHub is not alone in this transition. OpenAI and Anthropic have already moved enterprise customers to token-based billing. The consequences are visible at scale. Uber’s CTO recently told The Information that the company spent its entire AI budget for 2026 within the first few months of the year. Uber relies primarily on Anthropic’s Claude coding agents, and 11% of its code updates are now AI-generated. That kind of volume adds up fast under per-token pricing.

Outside pure software development, companies deploying AI automation for business processes should take note. Complex tasks that involve running large language models unsupervised for extended periods will likely face similar billing structures soon. The promised efficiency gains from AI in the workforce must now be weighed against rising vendor bills. A task that saves three hours of human labor might cost more in tokens than expected.

What This Means for the Future of AI-Assisted Coding

The era of free or flat-rate AI coding assistance is ending. Developers and businesses will need to think like cloud architects, monitoring usage, optimizing prompts, and choosing models based on cost efficiency. Tools that abstract away token tracking, like usage dashboards or budget alerts, will become essential. The winners in this new landscape will be those who can write efficient queries and leverage caching effectively.

GitHub’s move signals a maturing market. AI coding assistants are no longer experimental toys; they are production tools with real economic costs. The question now is whether the productivity gains justify the per-token spend for every query typed. For many teams, the answer will depend on how well they adapt to thinking in terms of tokens instead of requests. One thing is certain: the days of treating Copilot like an infinite resource are numbered, and developers who ignore their token diet may find their budgets depleted faster than their coffee mugs.