GitHub Copilot is transitioning to a usage-based billing model, effective June 1, 2026, shifting away from its prior premium request unit (PRU) system. This change is driven by Copilot's evolution from a simple editor assistant to an "agentic platform" capable of executing complex, multi-stage coding sessions and repository-wide tasks, which significantly increased its demand for compute and inference resources, rendering the previous fixed-cost model unsustainable.

Under the new model, all GitHub Copilot plans will bill based on monthly GitHub AI Credits. Usage will be calculated by consumption of input tokens, output tokens, and cached tokens, as determined by the public API rates of the underlying models. The base prices for all existing plans (e.g., Copilot Pro at $10/month, Copilot Business at$19/user/month) will remain unchanged. However, the existing "fallback experience" (which allowed continued usage with a lower-cost model when PRUs were exhausted) will be discontinued; future usage will be strictly governed by available AI Credits and administrator-set budget controls. Code completions and Next Edit suggestions will continue to be included without consuming AI Credits. Notably, Copilot Code Review will consume both GitHub AI Credits and GitHub Actions minutes, billed at the standard Actions per-minute rate.

Methodology of Usage Calculation:
The core methodology for billing shifts to a token-based consumption model. For each interaction with Copilot's AI capabilities, the system will track the number of tokens processed for input (prompts), generated for output (responses), and tokens that are cached. This token count is then multiplied by a specific "model multiplier" or "rate factor" ($\text{M}$) unique to the AI model being used (e.g., Opus 27x, Sonnet 9x, GPT 6x). The total cost for a session or period is calculated as:
$$\text{Cost} = \sum_{\text{sessions}} (\text{InputTokens} + \text{OutputTokens} + \text{CachedTokens}) \times \text{ModelAPIPricePerToken} \times \text{M}$$
This sum is then deducted from the user's monthly AI Credits. If the included credits are exhausted, additional usage can be purchased at the public rates, or constrained by set spending caps.

Impact on User Groups:

• Individual Users (Monthly Plans): Monthly Copilot Pro and Pro+ subscribers will automatically transition on June 1, 2026. Pro plans will include $10 of AI Credits per month, and Pro+ plans will include$39 of AI Credits per month.
• Individual Users (Annual Plans): Annual Pro or Pro+ subscribers will maintain their existing PRU-based billing until their current plan expires. However, from June 1, 2026, model multipliers for annual subscribers will increase significantly (e.g., Opus jumping from 3x to 27x, Sonnet from 1x to 9x), effectively increasing their usage cost for powerful models within the existing PRU framework. Upon expiration, they can convert to Copilot Free or upgrade to a monthly paid plan, with prorated credits for any remaining annual plan value if they switch early.
• Business and Enterprise Users: These plans maintain their per-user pricing ($19/user/month for Business,$39/user/month for Enterprise) and include respective AI Credits. A key change is the introduction of "pooled included usage" at the organizational level, allowing unused credits to be shared across the organization rather than being "stranded" per user. New "budget controls" will be available for administrators to set spending limits at the enterprise, cost center, or individual user level. Promotional included usage ($30/month for Business,$70/month for Enterprise) will be provided for June, July, and August 2026 to assist with the transition.

The market reaction indicates that many users perceive a significant decrease in the value proposition, especially for heavy users who previously benefited from highly subsidized inference costs. The change reflects a shift in GitHub's strategy from a "loss leader" model aimed at user acquisition to one focused on long-term sustainability and aligning pricing with the actual, increasing costs of advanced AI model inference, particularly for agentic workflows. This move is seen by some as the end of heavily subsidized AI compute and a push towards operational efficiency, potentially driving users towards direct API providers or alternative local models if the new pricing proves prohibitive for their usage patterns.