App teardown finds seven hidden Gemini Live models, including new 'Thinking' mode

Google looks like it’s getting ready to make Gemini Live feel less like one voice mode and more like a menu of specialized assistants. That’s the real story in this latest app teardown. Hidden code in the Android Google app points to seven unreleased Gemini Live model options, including a “Thinking” variant and personalization-focused builds that seem designed to use your history and app data. None of that means launch is guaranteed. But it does mean Google is actively wiring these choices into a consumer app right before I/O. ### What was actually found? The teardown centers on Google app version 17.18.22, where a hidden model selector for Gemini Live appears to list seven options that users can’t access yet. The reporting points to a much more granular setup than today’s simple Live experience, where Google mostly decides the model behavior for you behind the scenes. In plain English — Google seems to be preparing a chooser, not just a silent backend upgrade. (forbes.com) ### Why is “Thinking” the interesting one? Because Google has already defined “thinking” as a real model behavior in its developer stack. In Gemini’s API docs, thinking models spend extra compute on reasoning and multi-step planning before answering. Earlier teardown strings for Gemini Live describe “Live Thinking Mode” almost the same way — a version that “takes time to think” and gives more detailed responses. So this is not just branding fluff. It lines up with a real capability Google already exposes to developers. (forbes.com) ### What about the personalization models? Those may be the bigger product shift. Earlier code strings tied Gemini Live to “personal context,” past conversations, and connected Google apps. Another teardown described this as Personal Intelligence coming to Live — basically, the assistant remembering enough about your inbox, photos, habits, and prior chats to answer in a more tailored way. That would move Gemini Live closer to the assistant Google has been hinting at since Project Astra — less generic chatbot, more always-available helper. (ai.google.dev) ### Why hide seven models instead of one? Because voice assistants have conflicting jobs. Fast replies need low latency. Better reasoning needs more compute and more time. Personalization needs access controls and probably different privacy handling. A single “Gemini Live” button hides those tradeoffs, but a model picker lets Google offer modes for speed, depth, and context without pretending one model is best at everything. The developer side already reflects that split — Google lists separate Live, Flash, Pro, and thinking-capable families in the Gemini docs. (androidauthority.com) ### Does this mean these models will launch at I/O? Not automatically. APK teardowns catch work in progress, and Google often tests ideas that never ship. But the timing is hard to ignore. Google has already confirmed I/O 2026 for May 19–20 at Shoreline Amphitheatre, and it has explicitly said the event will feature AI breakthroughs and updates across products, from Gemini to Android. So even if these exact labels change, the direction of travel looks pretty clear. (ai.google.dev) ### What would this change for normal users? The simplest change is that Gemini Live could stop feeling one-size-fits-all. You might use a faster mode for casual back-and-forth, a thinking mode for harder questions, and a personalized mode when the answer depends on your stuff. That sounds small, but it changes the product from “talk to Google’s AI” into “pick the kind of help you want right now.” That’s a much more legible product strategy. (blog.google) ### What’s the catch? More capable modes usually mean more waiting, more complexity, or more data use. Thinking models are slower by design. Personalized models raise obvious privacy questions, especially if they pull from Gmail, Photos, and past chats. Google can make that useful, but it also has to make the boundaries obvious. Otherwise the smarter assistant starts to feel creepy fast. ### Bottom line? This leak matters less because of the number seven and more because of what the list implies. (forbes.com) Google seems to be breaking Gemini Live into distinct modes for speed, reasoning, and personal context. If that lands at I/O, Gemini Live stops being just a voice feature and starts looking a lot more like Google’s real consumer AI platform. (ai.google.dev)