320
Audio & Video Production306
Software Development226
Automation & Workflow203
Writing & Content Creation185
Marketing & Growth175
AI Infrastructure & MLOps143
Design & Creative150
Photography & Imaging141
Data & Analytics109
Voice & Speech123
Education & Learning119
Sales & Outreach112
Customer Support110
Research & Analysis84
GitHub says Copilot will move to AI Credits and token based charges for heavier use, while basic code suggestions stay included.
In short: GitHub says it will start charging Copilot users based on how much AI they actually use, starting June 1.
GitHub announced a new billing system for GitHub Copilot, its paid AI assistant for writing and reviewing code. The change starts on June 1.
Today, Copilot plans come with monthly “requests” and “premium requests.” GitHub says that system no longer matches its real costs, because a simple question can cost the same as a long, complex session where Copilot does a lot of work.
Under the new system, subscribers will get a monthly allotment of “AI Credits” tied to what they pay for their plan. If they go over, extra charges will be based on “tokens,” which are small chunks of text the AI reads and writes (like counting words, but in smaller pieces). Different AI models cost different amounts per token, so using a more powerful model can cost more.
GitHub says some everyday features will still be included without using AI Credits, like code completion and “Next Edit” suggestions. Copilot code reviews will add a different kind of cost, in the form of GitHub Actions minutes (time used on GitHub’s built-in automation service).
GitHub will also offer a “preview bill” tool before June 1, so customers can estimate what their current usage would cost under the new rules.
For individuals and teams, Copilot bills may become less predictable. It will be more like a phone plan that charges extra when you go over your monthly data, rather than a single flat fee. GitHub says it can no longer absorb rising AI running costs, especially from the heaviest users.
Source: Arstechnica