Google Launches Gemini 3.8 Live With Extended Thinking

Google has introduced two new live dialogue models, Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking, touting them as their most sophisticated live AI models to date. These models aim to enhance voice conversations with AI, making them more natural and powerful. Gemini 3.8 Live Extended Thinking is the more advanced model, intended for complex tasks that demand multi-step reasoning.

The new model can reason and speak simultaneously, enabling it to maintain a conversation while performing intricate tasks in the background. It can offer initial verbal responses like "Let me check that..." and provide real-time progress updates while completing multi-step workflows. The model is designed for features such as Gmail Live, Docs Live, and Keep Live.

Gemini 3.8 Live Extended Thinking has demonstrated impressive benchmark results, achieving an 82.6 score on Speech-to-Speech Quality Index, topping the overall rankings. It scored 68.6% on τ-Voice for agentic task completion and 35.1% on Sierra's τ-Voice banking benchmark. In reasoning, the model reached a remarkable 97.7% on Big Bench Audio. Moreover, the model offers competitive pricing compared to other frontier models.

On the other hand, the standard Gemini 3.8 Live model emphasizes conversational intelligence, fluid dialogue, and visual understanding. It can process visual inputs in near real-time while ongoing tasks continue in the background. This allows the model to acknowledge requests and continue the conversation while tasks are still being completed.

It can also understand and switch between 97 languages during a single conversation. Designed for scalability and cost efficiency, Gemini 3.8 Live currently holds second place in the Speech Agent Arena and powers the Search Live experience in AI Mode.