XNews

Google Releases Gemini 3.8 Live with Extended Thinking

Google Releases Gemini 3.8 Live with Extended Thinking

Google released Gemini 3.8 Live (cost-efficient, 97-language support, real-time multimodal) and 3.8 Live Extended Thinking (multi-step reasoning, synchronized thinking and speaking). Extended Thinking handles complex tasks with background tool calls and progress narration, usable as a tutor.

Key insights

Gemini 3.8 Live: cost-efficient real-time audio

Gemini 3.8 Live focuses on high-volume deployment with competitive cost. It supports automatic detection and smooth switching across 97 languages in conversation. It also offers near-real-time multimodal interaction (vision + speech) and background tool calling without disrupting chat. This model is aimed at high-concurrency, low-latency assistant and retrieval services.

Gemini 3.8 Live Extended Thinking: real-time reasoning with narration

The Extended Thinking version is designed for enterprise-grade complex tasks with multi-step reasoning and planning. It synchronizes thinking and speaking: when facing a complex problem, it naturally says things like 'let me check that' while executing multi-step tasks asynchronously in the background, updating the user on progress as if on a phone call. It also supports near-real-time visual understanding, e.g., looking at a chessboard or sketch.

Asynchronous tool calling and visual understanding

Both models feature async tool calls: the model can call external APIs, retrieve data, or book a reservation while front-end voice chat continues without interruption. Extended Thinking adds increased performance and precision for difficult tasks and narrates task progress to keep the conversation going. The models were demonstrated as a programming tutor.

Source posts