GitHub's shift from request-based billing to usage-based pricing for Copilot is hitting subscribers hard as the new system goes live. Users report burning through entire monthly AI credit allotments in single days, with some hitting 21% of their subscription cap after just basic queries. The sticker shock reveals how much GitHub was previously subsidizing heavy usage patterns.

Under the new model, paid subscriptions grant AI credits at $0.01 per credit, with the $10/month Pro plan offering 1,500 credits ($15 worth) and the $100/month Copilot Max providing 20,000 credits. Credit consumption varies dramatically based on model selection and token usage - simple prompts can cost 15-100 credits while complex ones burn 171-700 credits. The underlying LLM choice significantly impacts costs, with GPT-5.5 frontier models costing substantially more than nano variants.

Some users are adapting by making more deliberate AI interactions, like limiting themselves to focused changes or avoiding extended chat sessions that resend entire histories. However, many subscribers are threatening to cancel rather than pay for what were previously free heavy-usage patterns. The pricing change comes as Microsoft faces pressure to monetize AI services more effectively.

Alternatives like Deepseek are already attracting attention for efficiency - one user reports just 7 cents for 15 million tokens. This pricing shift signals a broader industry move toward usage-based models, where token-efficient models may gain competitive advantage as users become more cost-conscious about every prompt.