Two versions for different workloads

Google introduced Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking in September. The first is positioned for scaled, lower-cost voice interactions; the second aims at more complex tasks that require longer reasoning. Both are part of Google's push to make voice an interface for actual work rather than a turn-by-turn question box.

Google describes visual context, language switching and background tool calls. In the company's examples, the assistant can acknowledge a request, continue a conversation and report progress while external actions complete. Those capabilities are potentially useful for customer support, accessibility and hands-free workflows. They also make it more important to show the user which action the system is taking and when it is done.

Read the performance claims carefully

Google cites external and internal evaluations in its announcement, including speech quality and task completion results. The methods, evaluation conditions and real-world fit matter more than any one ranking. NEURA has not run independent tests of latency, error recovery or multilingual accuracy.

Google says the models are rolling out through its developer tools and consumer products, while enterprise access is in private preview or planned for some offerings. Readers should check the specific product and account they use before assuming a feature is available.

The next test

Continuous conversation is compelling when interruptions, background tool calls and visual context work together. The hardest measure is whether a user can trust the result of a multi-step task. Developers evaluating the models should test failure handling and confirmation flows alongside response quality.