Gemini Live: Google's 7 AI Voices, incl. "Thinking" Variant

Google is preparing a system for Gemini Live that will let users switch between seven different AI voices. This includes a "Thinking" variant designed for enhanced reasoning. Two of these models are already at the Release Candidate 2 stage. A hidden model selector was found in a recent Google App version, revealing these previously unreported AI options. The full list includes "Default," "Capybara," "Nitrogen," and a specialized personalization variant. "A2A" likely stands for Audio-to-Audio, meaning these models process speech directly. Tests show these models produce measurably different responses. The personalization variant remembered personal details shared earlier, unlike the current default Gemini Live model. This architecture allows Google to add or remove models without an app update. Separately, a user gained early access to "Gemini Omni," Google's new AI video model. Omni handled complex reasoning prompts with lifelike results, though some glitches appeared in more complex scenes. The computing cost for Omni is steep, with just two video generation requests using up most of a user's daily limit. This suggests why Google is adding explicit usage limits for Gemini. These discoveries point to a busy I/O 2026, as Google develops both a multi-model Gemini Live experience and advanced AI video generation. This means you could soon have more dynamic and personalized AI interactions.