Google DeepMind launches Gemini 3.8 Live and 3.8 Live Extended Thinking with focus on latency and tool calling

Google DeepMind featured Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking, two new versions of their AI model focused on real-time interactions. According to official announcement, the highlight is the combination of low latency, response fluidity and advanced tool call capabilities — the ability to trigger services, APIs and corporate systems in the middle of a conversation. The launch also includes the SynthID watermark and partnerships with companies such as Salesforce, ServiceNow and Genspark.

The movement consolidates a logic that has been established in the border models: to divide the experience into two ways. One quick, for continuous conversation; one slow, for problems requiring longer reasoning.

What changes on Gemini 3.8 Live and Live Extended Thinking?

The stated focus is short response time and fluidity — the feeling of talking to something that does not stop at each sentence. Live mode prioritizes the pace of interaction, while Extended Thinking accepts a longer processing time in exchange for more elaborate responses when the task requires careful analysis.

The novelty fits into a line Google was already building. At Google I/O 2026, the company showed Gemini 3.5 Flash, optimized precisely for agents and use on corporate platforms — read more on our coverage of Google launches Gemini 3.5 Flash optimized for agents and Omni. 3.8 Live takes a step forward by putting real-time interactivity at the center of the product, not as an accessory feature.

Why is the tool call the technical highlight of Gemini 3.8 Live?

Tool call (tool calling) is the mechanism that allows a model not only to respond, but to act: query a database, open a ticket, trigger a workflow or pull information from a CRM. The more fluid this capacity within a real-time conversation, the closer the assistant comes to a real operator.

That's where the ad talks to the ultra-low latency race. Infrastructure providers have been launching specific solutions for rapid inference, as shown in the coverage of joint AMD and Cerebras solution for ultra-low latency inference. Quick response is a prerequisite for any agent who needs to act while the user is still talking.

Gemini 3.8 Live Partnerships with Salesforce, ServiceNow and Genspark

To explore the new features, DeepMind has signed partnerships with companies such as Salesforce, ServiceNow and Genspark. They are names consistent with the proposal: Salesforce and ServiceNow dominate customer management systems and business workflows — territories where an assistant able to call tools has immediate use. Genspark operates in the field of agents and searches with AI.

What the announcement does not say is equally relevant: there is no detail of scope, deadlines, exclusivity or which products specific to each partner will receive the resources first.

SynthID and transparency in the content generated by Gemini 3.8 Live

The launch includes the SynthID watermark, Google technology that inserts an imperceptible signal into AI-produced content to allow later identification. In practice, it is a response to increasing pressure for traceability — whether a text, audio, or video came out of a model.

It's worth the warning. Watermark is not equivalent to infallible detection. It helps to verify origin, but depends on broad adoption and compatible tools for real everyday use.

What's the Gemini 3.8 announcement? Live still not responding

There are no comparative benchmarks, prices, general availability dates or published usage limits. It is also unclear in which regions Gemini 3.8 Live comes first. For those developing in Brazil, this means waiting for API documentation and pricing policy before planning integrations — but the message is already given: the 2026 dispute is less about "who responds better" and more about who responds quickly enough to act.

Frequently Asked Questions about Gemini 3.8 Live

What is Gemini 3.8 Live?

It is a version of Google DeepMind's AI model focused on real-time interactions, focusing on low latency, fluidity and ability to call external tools during conversation.

What is the difference between 3.8 Live and 3.8 Live Extended Thinking?

Live mode prioritizes response speed; Extended Thinking adds longer and deeper reasoning when the task requires more careful analysis.

What does the SynthID watermark do?

It inserts an imperceptible signal in AI-generated content to help identify its origin and enhance transparency — although it does not replace independent verification.