Overview of the Discovery
Recent findings in the Google App v17.18.22 reveal that Google is testing new AI models for Gemini Live, including a variant called “Thinking.” This hidden feature allows users to select from seven different AI models, each providing unique responses during voice interactions. The testing appears to be an internal tool ahead of Google I/O 2026, indicating that significant upgrades may be on the horizon for the voice-controlled chatbot.
Key Details
- Seven AI model options have been identified, including “Capybara” and “Nitrogen.”
- Four of these models can access user location for live weather updates, while three require users to provide their location.
- The “Capybara” model refers to itself as “Gemini 3.1 Pro,” suggesting enhanced capabilities over the standard model.
- The presence of a “Thinking” model indicates a focus on improving reasoning abilities for more thoughtful responses.
Importance of These Developments
The introduction of switchable models could greatly enhance the user experience for Gemini Live. Users may soon have the option to choose between fast responses and more thoughtful, slower replies. This flexibility could cater to different user needs, allowing for a more personalized interaction. As Google prepares for its upcoming event, these advancements highlight the company’s commitment to improving voice AI technology, potentially transforming how millions of users engage with their devices.
