GitHub Copilot Switches to Token-Based Billing

GitHub Copilot will transition to a token-based billing system starting June 1, 2026, replacing the current Premium Requests model. The shift changes how the AI coding assistant calculates usage, moving from a system where every query counted as a single request to one where costs are determined by the volume of data processed.

Under the existing Premium Requests system, users were charged per query regardless of the complexity of the prompt or the length of the generated code. The new model will instead base charges on the number of tokens processed, with rates varying based on the specific AI model selected, the volume of input and output, cache size, and the specific features utilized during a session.

The Mechanics of AI Credits

While subscription prices will remain the same, the way those subscriptions are applied will change. Users will now receive AI Credits instead of a fixed number of queries. For example, a Copilot Pro subscriber paying $10 per month will be allocated 1,000 AI Credits, with GitHub valuing each individual credit at one cent.

In the context of large language models, a token is a basic unit of text, roughly equivalent to three-quarters of a word. To illustrate the scale of this usage, a codebase containing 10,000 expressions, statements, and variable names typically translates to approximately 12,000 to 13,000 tokens. Both the input prompts provided by the developer and the output generated by the model contribute to the total token count.

Impact on Developer Workflows

The rate at which credits are consumed will depend heavily on the complexity of the task and the sophistication of the model used. More advanced models will deplete credits more quickly than less capable versions. Similarly, complex workflows—such as those involving multiple agents operating across large codebases—will consume credits at a faster rate than simple, single-turn queries.

Certain features will remain exempt from this new billing structure. Code completions and Next Edit suggestions will not be affected by the transition to token-based billing and will continue to be provided for free.

For many users, this change represents a reduction in the effective subsidy provided by GitHub. Under the previous model, users typically utilized between three and eight times the token value that their subscription cost actually covered, with GitHub absorbing the remaining expense. This subsidy will end on June 1, 2026.

Developers who primarily perform simple queries are unlikely to exceed their monthly credit limits. However, those managing complex workflows or working within extensive codebases may find their credits depleted more quickly and may need to purchase additional credits to maintain their current level of usage.

Industry Alignment and Predictability

GitHub’s move aligns its pricing strategy with adjustments already made by other major AI providers, including OpenAI and Anthropic, specifically for their enterprise customers. Token-based billing allows AI providers to gain more direct insight into the actual compute costs associated with specific users and tasks.

Despite the alignment with industry standards, the shift introduces a new layer of unpredictability for the end user. While the previous fixed monthly fee provided a stable cost regardless of usage volume, the token-based system makes monthly expenses more volatile based on the intensity of the developer’s work and the models they choose to employ.

As of May 2, 2026, GitHub has not yet published a detailed breakdown of the specific per-token rates for each of its supported model tiers.