Google introduced Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking, which keep conversations going while handling tool and API calls in the background. The standard model focuses on scalability and cost efficiency, while Extended Thinking focuses on complex, multi-step reasoning. The models offer language detection, switching, voice generation, and visual processing in 97 languages. Demos showcased examples involving workplace adaptation, chess, React development, and reservations. Extended Thinking outperformed its rivals with a score of 82.6 on the Speech to Speech Quality Index, while the standard model placed second in the Speech Agent Arena. Available through the Gemini API and Google AI Studio, the models are in private preview in Gemini Enterprise and are also coming to Search Live, Gemini Live, and Workspace. The voices include SynthID. The standard model’s per-minute price is $0.005 for input and $0.018 for output.
Why it matters
Keeping conversations uninterrupted while tools and APIs are used in the background could make voice interactions such as reservations, software development, and enterprise workflows more seamless. The two-version structure allows developers to choose between cost and scalability for high-volume use and reasoning capacity for multi-step tasks. Bringing multilingual and visual capabilities to Search, Workspace, and Gemini products shows that the development concerns not only API users but also a broader audience using Google services. Although the announced performance metrics provide a basis for comparison, they do not by themselves demonstrate the demos’ consistency under real-world conditions. In addition, the disclosure of pricing only for the standard version leaves the cost of using Extended Thinking unclear.
Background
Google is not a new name in the FikirPilot archive: over the past 90 days, we have published 41 news stories mentioning this name; the latest was dated September 15, 2026.