In the ever-evolving world of AI, a recent discovery has shed light on Google's potential plans for its voice-controlled chatbot, Gemini Live. A hidden menu within the Google App has revealed a glimpse into the future of AI-powered voice assistants, and it's an exciting prospect that raises many intriguing questions.
Unveiling the Hidden Models
The revelation of seven new AI models, codenamed 'Capybara,' 'Nitrogen,' and others, suggests that Google is experimenting with various approaches to enhance Gemini Live's capabilities. These models, accessible through a hidden selector, offer distinct functionalities, as evidenced by my own testing.
Personalization and Thinking Variants
One of the most fascinating aspects is the presence of a 'Thinking' variant. This suggests an AI model with enhanced reasoning abilities, a significant step forward in natural language processing. Additionally, a 'Personalization' model hints at a more tailored and adaptive assistant, capable of understanding and responding to individual user preferences.
What Makes This Discovery Significant
Currently, Gemini Live operates with a single model, Gemini 3.1 Flash Live, designed for processing raw audio and video streams. The existence of multiple new models indicates that Google is actively exploring alternatives, potentially paving the way for a more diverse and powerful voice assistant ecosystem.
Implications for Users
If Google were to introduce model selection for Gemini Live, it could revolutionize the way users interact with their voice assistants. It would enable users to choose between faster, more responsive models or opt for more thoughtful, deliberate responses, catering to different preferences and use cases.
The Mystery of 'Capybara' and 'Nitrogen'
The codenames 'Capybara' and 'Nitrogen' remain shrouded in mystery, with no prior documentation to offer clues. However, the presence of terms like 'Rev25' and 'Exp' suggests an extensive development process, with multiple revisions and both stable and experimental versions under testing.
A Step Towards Customization
The ability to add or remove models without an app update showcases Google's commitment to flexibility and customization. This dynamic approach allows for rapid iteration and the potential to offer users a personalized experience without the need for frequent app updates.
Testing Reveals Functional Differences
My testing confirmed that switching models indeed produces different behaviors. Four models were able to access user location for live weather updates, while others required manual input, demonstrating a clear distinction in their capabilities.
The Future of Voice AI
As we approach Google I/O 2026, it's evident that Google is road-testing a range of voice AI capabilities. The infrastructure is in place, and the list of models is growing, suggesting that model selection could be a feature on the horizon for Gemini Live users.
In conclusion, this hidden menu offers a tantalizing glimpse into the future of voice-controlled AI. It raises questions about the potential for more powerful, personalized assistants and highlights Google's commitment to innovation in this space. As we await further developments, one thing is clear: the world of AI is about to get even more exciting.
