Google Testing 7 Hidden Gemini Live AI Models Ahead of I/O 2026

What Are the Hidden Gemini Live Models?

A hidden model selector discovered in Google App v17.18.22 reveals seven previously unknown AI voice models being tested internally. Each model — codenamed things like "Capybara" and "Nitrogen" — produces measurably different responses in testing.

Why This Discovery Matters

Google has never offered multiple voice AI models to consumers. This hidden selector suggests Google I/O 2026 could introduce a choice-based voice AI experience, letting users pick models optimized for different tasks — weather queries, personal assistance, or deep reasoning.

What Each Model Does Differently

Testing reveals distinct capabilities across the seven models. Four can access user location for real-time weather data; three cannot. One model, "Capybara," identifies itself as "Gemini 3.1 Pro" rather than the standard Flash Live model. Others show different memory retention within conversations and varying accuracy levels.

The Bigger Picture for Voice AI

This move signals a shift from one-size-fits-all voice assistants to specialized AI experiences. Instead of a single model trying to do everything well, users may soon choose models tuned for their specific needs — a philosophy that could reshape how we interact with AI daily.

FAQ

Q: Can I access these hidden models now? A: No. They're behind a server-side flag for internal testing. Expect an official reveal at Google I/O 2026.

Q: Will different models cost different amounts? A: Google hasn't announced pricing. The current Gemini Live is free, and most analysts expect a freemium model with advanced options.

Q: Why would I want different voice AI models? A: Different tasks need different strengths. A quick weather check needs speed; a brainstorming session needs depth and memory. Specialized models serve each better.

Stay ahead of the AI curve. Follow @AiForSuccess for daily insights.