Google has introduced new rules for measuring and managing Gemini AI usage, shifting from counting the number of requests to assessing the computational power required for each task. This change affects all Gemini AI tiers, including Free, Plus, Pro, and Ultra. Users may now encounter usage limits more quickly when working with complex prompts, such as video generation or advanced coding tasks. The company explained that this approach helps manage resource costs and ensures fair usage across its data centers. However, the new system is less predictable for users, who are now less certain about when they will hit their limits. The change also means that simple tasks like asking for a weather forecast will use fewer resources than more complex operations. Source: wired

The new pricing structure offers four subscription tiers in the US: AI Plus ($8/month), AI Pro ($20/month), and AI Ultra ($100 or $200/month). Each plan provides progressively higher usage limits, with AI Ultra offering either 5x or 20x more than AI Pro, depending on the payment level. Free users receive standard limits, while Plus and Pro users get 2x and 4x those limits, respectively. All users can access the same Gemini AI models, including Flash-Lite, Flash, and Pro, though higher-tier models use more resources and offer larger context windows. The context window size, which determines how much text can be processed in a single conversation, ranges from 32K tokens for free users to a million tokens for Ultra subscribers. Source: wired

Google’s support documents state that access to AI features may be limited or changed based on testing, experimentation, or availability. This means users might experience varying levels of AI functionality on different days. Additionally, if users exceed their weekly or five-hour reset limits on paid plans, they will be demoted to the most basic AI model until the next reset. Google also warns that usage limits could change without notice due to resource constraints, with free users potentially affected first. Source: wired

Source: wired