GitHub Copilot is set to implement a significant pricing model change: starting next month, its AI services will transition to a per-token charging system. A token is typically described as representing approximately three-quarters of a word. This implies that if a Large Language Model (LLM) were to analyze a text of 10,000 words, it would consume around 12,000-13,000 tokens.
For developers, if Copilot examines a code base comprising 10,000 'words' (e.g., expressions, statements, variable names, functions) for tasks like refactoring or bug-hunting, a single query could consume 12,000-13,000 tokens from their monthly allotment. Both prompt inputs and Copilot's outputs will count towards token usage.
The upcoming pricing tiers will maintain current monthly subscription levels, but instead of a fixed number of queries, users will receive 'AI Credits' of equivalent value. For instance, a base-tier Copilot Pro subscriber ($10 per month) will receive 1,000 credits, with GitHub stating that one AI Credit is currently worth one US cent.
The number of tokens each credit purchases will vary based on several factors: the specific AI model utilized, the input/output mix, the size of the cache (data held in the LLM's memory for context), and the feature requested. This means developers primarily using simple queries might not need to purchase additional credits monthly. Conversely, complex multi-agent queries involving extensive code bases will deplete AI Credit accounts more rapidly. Furthermore, queries directed to the most advanced frontier models will incur higher costs than those to less powerful models.
It's important to note that GitHub's pricing adjustments do include some compensatory benefits: core functionalities such as code completions (similar to a phone's auto-complete) and Next Edit suggestions will continue to be offered free of charge to users.