Google announced a significant overhaul of the Gemini app for users on the free plan. The changes take effect immediately, altering how users select AI models and adding new „thinking” or „effort” levels to the experience. Free users no longer have the option to manually choose a model; the picker now defaults to „Auto” with Gemini 3 models. The update also introduces a tiered thinking system that allows the AI to decide how much computational effort to apply.
The decision follows last week’s rollout of Gemini 3, which introduced higher‑capacity models and improved contextual understanding. By restricting manual model selection for free users, Google aims to simplify the interface and encourage the use of its most efficient, cost‑effective model. At the same time, the new thinking levels let the AI choose between quick, lightweight responses and more thorough, resource‑intensive answers. This dual change is intended to balance performance with cost savings for the company while giving users finer control over response depth.
Under the new system, free users are automatically directed to the Gemini 3 models. The „Auto” setting evaluates the complexity of a query and selects the most suitable model from the Gemini 3 family. This eliminates the need for users to manually toggle between versions, reducing confusion and streamlining the experience. The Gemini 3 models are optimized for speed and efficiency, ensuring that free users receive quick answers without compromising quality. The change also aligns with Google’s broader strategy to standardize the free tier, making it easier to manage infrastructure and cost.
Google’s new „thinking” levels are a form of adaptive computation. When a user submits a question, the AI can choose to spend more or fewer resources to generate a response. Higher thinking levels trigger deeper analysis, longer context windows, and more sophisticated Lower levels favor speed and lower computational cost. This feature is particularly useful for complex queries that require multi‑step For simpler questions, the system conserves energy, keeping response times short. The addition of these levels represents a step toward more efficient AI usage, potentially reducing operational costs for Google while offering users a more nuanced experience.
Free users will notice a smoother, more consistent interface. The removal of manual model selection reduces decision fatigue, especially for those who are not familiar with AI model differences. The new thinking levels give users indirect control over response depth without needing to understand technical parameters. However, some power users may miss the ability to pick a specific model or adjust settings manually. Overall, the changes are designed to make the Gemini app more accessible and cost‑effective for the broadest audience.