GitHub has announced significant changes to Copilot’s individual subscription plans. New sign-ups for the Pro, Pro+, and Student plans are being suspended for the time being. At the same time, usage limits are being tightened, and the range of AI models is changing.
According to the company, these measures are necessary to maintain service stability for existing users. The changes follow a rapid increase in Copilot usage, particularly driven by agent-based workflows. In these workflows, AI agents perform long-running, often parallel tasks that require more computing power than the original subscription structure was designed for. GitHub has observed that this leads to increased strain on the infrastructure and more frequent instances of users hitting their limits.
To alleviate this pressure, the company has decided to temporarily limit new sign-ups. At the same time, existing limits will be enforced more strictly. Users with a Pro subscription can upgrade to Pro+, which offers higher limits. Additionally, visibility into usage will be improved through notifications in Visual Studio Code and the Copilot CLI when users approach a limit.
In addition to the limits, the availability of AI models is also changing. Certain Opus models are being removed from Pro subscriptions and will become partially exclusive to Pro+. Older variants are being fully phased out, shifting functionality to more expensive subscriptions.
GitHub acknowledges that the changes have an impact and states that the balance between performance, cost, and scalability is under pressure. In some cases, individual requests now cost more than users pay. The measures are intended to prevent a deterioration in service quality.
According to reports by The Register, however, there is more at play than just product optimization. The publication outlines a broader context in which AI services are hitting infrastructural limits. Demand for computing power is growing faster than available capacity, affecting the entire sector. The publication also points to previous interventions, such as discontinuing free trial periods due to abuse.
Structural problem in pricing
In addition, The Register emphasizes that the current pricing model is under pressure. Some intensive AI requests reportedly cost the company more internally than what users pay through their subscriptions. This points to a structural pricing issue and could lead to further adjustments, such as a shift toward usage-based pricing.
The platform also recently introduced weekly usage limits in addition to session limits. While session limits are intended to manage peak load, weekly limits focus on total token consumption. Long-running, complex tasks in particular can quickly drive this up.
Users who reach limits can continue to use Copilot in a modified form, for example via automatic model selection. Full access to specific models returns after the usage period is reset. For those for whom the restrictions are too severe, subscription cancellation is possible with a refund for recent charges.