2026-09-16 · Fuente：SiliconANGLE
Google's newest speech models, Gemini 3.8 Live and Live Extended Thinking, target the latency problem in voice-based AI agents with near-real-time reasoning and simultaneous speech-and-thought processing.
Google LLC has unveiled two new speech models, Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking, positioning them as the company's most advanced voice processing models released so far. Announced today, the pair is aimed squarely at one of the most persistent obstacles in conversational artificial intelligence: the latency that keeps voice-based AI agents from feeling responsive and natural in live dialogue.
The headline capability is near-real-time reasoning. Google bills the models as capable of simultaneous speech-and-thought processing — reasoning that unfolds while speech is being produced, rather than in the listen-then-think-then-speak sequence that has historically characterized voice assistants. The Gemini 3.8 Live Extended Thinking variant signals a second mode oriented toward more deliberate reasoning, extending the thinking-model approach into Google's voice lineup under the flagship Live branding.
Latency has long been the defining constraint for voice-based AI agents. Users judge a voice interface in fractions of a second: pauses between a question and a response break the illusion of conversation in a way that delays in text chat never do. Pipelines that chain speech recognition, a reasoning model and speech synthesis compound delay at every stage. A model that reasons while it speaks attacks that bottleneck at its root, which is why Google describes the launch as an effort to address the latency problem associated with voice-based agents.
For enterprises evaluating voice agents for customer-facing workloads, the decisive questions now turn to availability, pricing and measured latency gains at production scale. Even so, the launch reads as a clear statement of direction: Google is framing real-time reasoning, rather than speech quality alone, as the benchmark that will matter for the next generation of voice-based AI agents, with Gemini 3.8 Live and its Extended Thinking sibling carrying that argument to market.
Etiquetas: Google · Gemini 3.8 Live · Gemini 3.8 Live Extended Thinking · Voice AI · Real-Time Reasoning · Latency
萬安算交所 AIXX · AIXX · Exchange verificado · Ir al mercado