According to foreign media reports, Google is making a significant adjustment for Gemini's free users: free users will no longer be able to manually switch between different models, with the model selection interface defaulting to "Auto" mode, where the system decides which model to invoke based on the question type. This change began rolling out on October 9, and most free accounts have now started seeing the related adjustments.
Google Gemini's free users to face model selection restrictions
According to Google's official help documentation, Auto mode allocates models based on the specific needs of each query. Most routine questions will be handled by the faster Flash-Lite, while tasks involving more complex reasoning may invoke Flash or even Pro models. Therefore, this adjustment does not mean free users completely lose access to higher-tier models, but users will no longer be able to actively specify models as before; which model is actually invoked will be primarily determined by the system.
Google also stated that if users disable the smart model selection feature, all conversations on free accounts will use Flash-Lite. This means that for users who need to reliably invoke specific models for programming, complex analysis, or other professional tasks, the new free-tier usage approach will bring certain limitations.
In addition to the model permission changes, Google is also rolling out new thinking intensity options for Gemini, including three levels — low, medium, and high — replacing the previous Extended option. Users can adjust the reasoning intensity the model applies based on task requirements; higher levels promise more in-depth answers but will also consume more usage quota. The relevant options are rolling out gradually.
In fact, Google began adjusting Gemini's usage quota in May 2026, switching to consumption calculated based on factors such as prompt complexity, features used, and conversation length. Quota refreshes every 5 hours, with a weekly cap also in place. Paid users receive higher usage quotas, while complex reasoning, deep research, and multimedia generation features consume quota more quickly.