AI costs how much? GitHub Copilot users react to new usage-based pricing system.

GitHub has implemented a new usage-based pricing model for its AI-powered Copilot service, replacing the previous request-based billing system. Under the new plan, users receive a monthly allotment of AI credits, with one credit equaling one cent of usage. Subscription tiers now include varying credit amounts: $10/month for 1,500 credits, $39/month for 7,000 credits, and $100/month for 20,000 credits. Many users have expressed surprise and concern on social media and forums as they find their typical usage rapidly depleting their monthly credit limits, sometimes within a single day. Previously, Copilot subscribers were charged based on a fixed number of “requests” and “premium requests” determined by their payment tier, which did not differentiate between short queries and extensive coding sessions. GitHub acknowledged that this system led to the company absorbing significant costs, especially for heavy users. Estimates shared by users suggest that their prior monthly usage could now cost thousands of dollars under the new pricing, highlighting the extent of GitHub’s previous subsidies for power users. The cost per prompt varies depending on the AI model selected, with cheaper models like OpenAI’s GPT-5.4 nano costing roughly $1.25 per million output tokens, while more advanced models such as GPT-5.5 can cost up to $30 for the same output. Users relying on Copilot’s “Auto” mode, which automatically selects the model, have reported unexpectedly high charges due to the system switching to more expensive models for simple tasks. For example, a basic prompt to build a Minesweeper game using Claude Haiku 4.5 consumed about 94 credits, illustrating how quickly costs can accumulate even for straightforward projects. This shift to usage-based pricing marks a significant change in how AI coding tools are monetized, potentially impacting developers who depend heavily on Copilot for their workflows. It underscores broader industry challenges in balancing affordable access to AI-powered services with the high computational costs of running advanced models. Users and organizations will need to carefully monitor and manage their AI usage to avoid unexpected expenses under the new system.