GitHub Copilot restrictions are being implemented, possibly due to a gradual shift to a token-based billing model following a nearly doubling of weekly operating costs starting in January 2026.
On April 20, 2026, GitHub announced that it would be suspending new subscriptions for its personal GitHub Copilot plan, tightening usage restrictions, and reviewing the available model. GitHub stated that 'these measures are in order to maintain a stable service provision system for existing users,' but other reports suggest that 'the weekly operating costs of GitHub Copilot have nearly doubled since January 2026, and Microsoft is considering a shift to token-based billing in the future.'
Changes to GitHub Copilot plans for individuals - GitHub Changelog
Changes to GitHub Copilot Individual plans - The GitHub Blog
https://github.blog/news-insights/company-news/changes-to-github-copilot-individual-plans/
Exclusive: Microsoft To Shift GitHub Copilot Users Token-Based Billing, Tighten Rate Limits
https://www.wheresyoured.at/news-microsoft-to-shift-github-copilot-users-to-token-based-billing-reduce-rate-limits-2/
This official announcement reveals that new registrations for Student, Pro, and Pro+ plans will be discontinued. GitHub explains that this measure is to prioritize the quality of service for existing paying users. New registrations for the free Copilot Free plan will continue, and existing users will still be able to change plans. However, usage limits for individual plans will be stricter, and warnings will be displayed in VS Code and Copilot CLI when approaching the limit.
Furthermore, the available model configurations will also be reviewed. Opus-based models will no longer be available in the Pro series, while Opus 4.7 will continue to be available in the Pro+ series, although Opus 4.5 and Opus 4.6 will be gradually removed.
GitHub cites the significant change in Copilot's computational demand due to the proliferation of agent-based workflows as the reason for these increased restrictions. They explain that the increase in long-running, parallel sessions has led to the use of more computing resources than anticipated in the original pricing structure, and it has become common for costs to exceed plan fees with just a few requests.
According to GitHub, Copilot currently has session-based and weekly limits, both of which are determined based on token consumption and a multiplier per model. The session limit is primarily intended to prevent overload during peak times, while the weekly limit was introduced to mitigate the high costs caused by long-running, parallel requests.
Furthermore, usage limits and the number of premium requests are treated separately. Therefore, even if you have remaining premium requests, restrictions will be imposed if your token consumption reaches the limit within a certain period. However, even if you reach the weekly limit, if you have remaining premium requests, you can continue using 'automatic model selection,' and manual model selection will become active again after the weekly period reset.
GitHub is suggesting ways to make it harder to reach the limit, such as using models with lower scaling for simpler tasks, improving task efficiency by using VS Code and Copilot CLI's plan mode, and upgrading to Pro+ for Pro users who need a higher limit. They have also announced a mitigation measure for users: if the changes don't suit them, they can cancel their Pro or Pro+ subscription and will not be charged for April usage, and they will be eligible for a refund if they contact GitHub support between April 20th and May 20th.
Meanwhile, Ed Zitron, who runs the newsletter 'Where's Your Ed At,' reports from internal Microsoft documents that 'Microsoft is considering eventually switching Copilot's billing method to token-based billing, and this has become a higher priority in recent months.' This is because the weekly operating costs of GitHub Copilot have nearly doubled since January. It seems that in the future, users will be charged based on the amount of tokens they consume, that is, closer to the computing resources they actually use, but Zitron says that the specific start date is unknown, and the official announcement so far has only been the suspension of new registrations and tightening of restrictions.
Related Posts:
in AI, Web Service, Posted by log1i_yk