Gemini Live Upgrades: New ‘Thinking Mode’ & Experimental Features
▼ Summary
– The current Gemini Live feature is powered by the Gemini 2.5 Flash model, but significant upgrades are indicated by code in a recent Google app beta.
– Code analysis reveals upcoming “Labs” features for the Gemini Android app, including a “Live Thinking Mode” for more detailed, considered responses.
– Another planned Labs feature, “Live Experimental Features,” includes multimodal memory, better noise handling, and personalized results from Google apps.
– Additional Labs capabilities in development are “UI Control,” allowing an agent to control the phone, and “Deep Research” for complex tasks.
– These discovered features are based on decompiled app code and may not all be released, as Google’s final plans can change.
The Gemini app for Android is poised for significant enhancements, with new experimental “Labs” features uncovered in the latest beta code. These upcoming capabilities suggest Google is preparing to upgrade the core technology behind Gemini Live and introduce powerful new functionalities for users who opt into testing them. While these features are still in development and not guaranteed for public release, they provide a compelling glimpse into the assistant’s potential future.
Code within the Google app version 17.2 points to four main Labs features. The first is “Live Thinking Mode,” described as a version of Gemini Live designed to take more time to process and deliver more detailed, thoughtful responses. This strongly indicates the feature will leverage more advanced AI models, likely moving beyond the current Gemini 2.5 Flash foundation. The second is “Live Experimental Features,” which bundles several cutting-edge abilities. These include multimodal memory, improved background noise handling, the capacity to respond to what it sees through the camera, and personalized results drawn from a user’s connected Google apps.
These two Labs specifically hint that the Gemini Live experience will soon be powered by the newer Gemini 3 model family. The “Thinking Mode” could utilize either the dedicated Thinking or Pro variants for its in-depth answers. Many of the experimental features, such as accessing personal data from Google apps, are already available in the chat experience with Gemini 3 Flash and Pro models. The visual response capability may be related to Project Astra, Google’s initiative for a more visually-aware AI assistant.
Two additional Labs features were also identified. One is “UI Control,” framed as an agent that can control a phone to complete tasks. The other is “Deep Research,” which allows users to delegate complex research tasks to the AI. The UI Control feature aligns with long-standing expectations for “Gemini Agent” to arrive on Android as part of broader computer use functionalities, though the exact implementation of Deep Research remains less clear. Together, these features paint a picture of a more capable, proactive, and integrated AI assistant that can handle a wider array of tasks directly on a user’s device.
(Source: 9to5 Google)
