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

GitHub Copilot users are currently experiencing "sticker shock" as a new usage-based pricing model goes into effect. This change, announced in April, has replaced the previous request-based billing system, leading many users to quickly burn through their monthly AI credit allotments. Reports from social media and forums indicate that even a few hours of AI usage can consume a significant portion of their monthly caps, with some users exhausting their quota in less than a day.

Under the new system, paid Copilot subscriptions now grant users a specific number of AI "credits" each month. One credit is equivalent to $0.01 of usage, with bonus credits provided depending on the subscription tier. For example, the $10/month Pro plan includes 1,500 credits, while the $100/month Copilot Max plan offers 20,000 credits.

The cost per prompt is determined by the number of input and output tokens used and the rates of the underlying large language model. This means pricing can vary significantly based on the chosen model; a simple query using an expensive model can rapidly deplete credits. Users relying on "Auto" mode to select models are particularly cautioned, as it may inadvertently switch to pricey options for basic tasks.

Examples of rapid credit consumption are abundant. A single complex prompt might use 171 credits, while a few prompts could consume 700. Even simple tasks like "generating a small plan" reportedly cost 100 credits, and one user consumed 840 credits during initial cautious testing. These high costs are prompting many to consider canceling their subscriptions or seeking alternative AI coding solutions.

However, some users have adapted to the new model by carefully managing their AI usage. One coder reported using only 161 credits during a productive day by focusing on deliberate changes. Another noted that continuing long chat sessions is now more expensive due to input tokens being used for context, suggesting a need for more efficient AI interaction.

While some users are leaving for services with more generous limits, it is anticipated that usage-based pricing may become a standard across the AI industry. In this scenario, efficiency in token usage will be crucial. Discussions have already begun about integrating more cost-effective LLMs, such as Deepseek, into development environments to minimize expenses. The future may see users not only paying for what they get but also carefully considering how they get it to control costs.