Key facts
- GitHub Copilot transitioned from request-based to token-usage billing on June 1.
- Some users report significantly higher projected AI bills under the new pricing.
- The new system converts usage costs into AI credits, with 1 credit equaling $0.01 of AI usage.
- GitHub cited unsustainable inference costs for advanced reasoning models and agentic workflows.
- Users can manage costs by avoiding lengthy chat sessions or summarizing long conversations.
GitHub Copilot has implemented a new token-usage-based pricing model, replacing its previous request-based system. This change, effective June 1, has led to concerns among some users about significantly increased costs. Power users have reported projected monthly bills hundreds of dollars higher than before, with some seeing their costs jump from around $45 to over $750. The new policy converts AI usage, based on the model and tokens consumed, into AI credits, where each credit is equivalent to $0.01 of AI usage. Subscribers receive a base allotment of credits, with bonus credits depending on their subscription tier. GitHub stated that the previous model was unsustainable due to rising inference costs associated with advanced AI models and agentic workflows, where complex tasks could incur the same cost as simple queries. Some users have criticized the change, comparing it to how ride-sharing services initially subsidized costs before increasing prices. Others defended the new pricing, attributing it to the underlying API costs of the AI models or suggesting users learn to code manually or optimize their usage, such as by avoiding summarizing long chat sessions. Analysts predict that more companies will adopt consumption-based pricing for AI services as compute demands increase. GitHub has referred users to an FAQ and changelog for further details on the pricing adjustments.