Google DeepMind Ships Gemini 3.8 Live With Extended Thinking Mode

Google DeepMind released Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking, two entries in its flagship model line that pair real-time interaction with an optional slower reasoning path. The company announced both on its own channels, positioning the pair as a single launch rather than separate products. The move keeps Google's live-capable models on a faster update cadence as rivals crowd the same territory.

The naming tells most of the story. 'Live' signals low-latency, streaming interaction across audio, video and text, the mode Google has been building since its first Live models. 'Extended Thinking' is the same base capability routed through a longer internal reasoning process before the model answers. Putting both under one version number suggests DeepMind wants developers to treat fast and slow inference as a dial rather than a choice between two unrelated systems.

That matters because the two modes solve different problems. A voice assistant handling a live camera feed cannot afford to deliberate for seconds; a coding or analysis task often can. Shipping both at once means a single integration can switch between them depending on the request, which cuts down on the model-juggling that teams building agent products have complained about. Developers get one API surface and one set of behaviors to tune.

Google has not published benchmark tables, context limits, pricing or regional availability in the announcement itself, so the practical gap between the standard Live model and the Extended Thinking variant is still unquantified. Those numbers will decide whether the reasoning mode is a genuine upgrade or a modest reroute. Until they land, the launch is a signal about direction more than a measurable step.

The competitive read is straightforward. OpenAI, Anthropic and a growing list of open-weight providers all offer some form of test-time reasoning, and live multimodal interaction has become table stakes for anything pitched as an assistant. What Google is selling here is the combination, plus the distribution that comes with its own products and cloud platform. That reach is the part competitors cannot copy quickly.

What to watch next is whether DeepMind publishes latency and cost figures for Extended Thinking, since those two numbers determine if the mode is usable in production or only in batch jobs. Also worth tracking is which Gemini surfaces get 3.8 Live first, and whether the older Live models are deprecated or left running alongside. The answers will show how serious Google is about making the fast-slow switch a default pattern for builders.