Google teases Gemini live models

Google looks ready to make Gemini Live less like one voice mode and more like a menu of specialized assistants. That’s the real signal heading into Google I/O 2026. A teardown of the Google app surfaced seven hidden Gemini Live model options, while Google has separately started teasing an Android-focused event on May 12 ahead of I/O on May 19 and 20. Put those together and the picture is pretty clear — Gemini is becoming a deeper part of Android, not just a chatbot floating above it. (forbes.com) ### What actually showed up? The new bit is inside Google app version 17.18.22. In that build, a hidden model selector for Gemini Live appears to list seven voice-conversation models, including variants tied to “Thinking” and personalization. None of this means all seven are launching next week. But it does mean Google is actively wiring Gemini Live for multiple modes instead of one default experience. (forbes.com) ### Why does “Gemini Live” matter here? Gemini Live is Google’s real-time spoken assistant layer — the part meant to handle back-and-forth voice conversations, not just one-shot text prompts. If Google gives that layer multiple models, it can start routing different jobs to different behaviors. One mode could answer quickly. Another could reason longer. Another could lean harder on your personal context. That is a much bigger change than adding a new voice or a prettier UI. (forbes.com) ### What’s the point of a “Thinking” model? Basically, Google seems to be borrowing the same playbook the rest of AI is moving toward — fast models for casual use, slower reasoning models for harder tasks. A “Thinking” label suggests Gemini Live may get a mode that pauses longer and does more deliberate reasoning before answering. That matters for spoken assistants because voice has usually favored speed over depth. Google looks like it wants both. (forbes.com) ### Why bring personalization into Live? Because voice is where a generic assistant starts to feel dumb fastest. If Gemini Live knows your habits, apps, preferences, and maybe ongoing tasks, it can answer in a way that feels continuous instead of starting from zero every time. The catch is privacy and trust — the more personal the assistant gets, the more Google has to show users what data is being used and how much control they have over it. The teardown only shows the direction, not those guardrails yet. (forbes.com) ### How does Android fit into this? Google has already confirmed The Android Show: I/O Edition for May 12 at 10 AM PT, a week before I/O begins on May 19 in Mountain View. That split matters. It suggests Google wants Android announcements to stand on their own, while the main conference handles the broader AI and developer story. If Gemini Live is getting new model tiers, Android is the obvious place to show how they surface on phones and other devices. (androidauthority.com) ### Is this just leak-season noise? Not really — but it is still pre-release evidence. APK teardowns show code paths and hidden UI, not shipping promises. Google could rename features, limit them to testers, or hold some back. Still, teardowns are useful when they line up with official scheduling and broader product direction, and that’s what’s happening here. Google is openly telling people Android has a big year ahead, while the app code hints that Gemini Live is one place where that gets real. (androidauthority.com) ### So what should you watch for next? Watch for Gemini Live to become more explicitly mode-based at The Android Show or I/O — faster mode, reasoning mode, and some kind of personal mode. Also watch for tighter Android hooks, because the whole point of this shift is that Gemini stops being just an app and starts acting like the intelligence layer across Google’s devices. That’s the bigger story. (forbes.com)