Google launches Gemini 3.8 Live models with background reasoning for real-time voice AI

Google has launched Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking, a pair of live-dialogue models built to hold real-time voice conversations while reasoning and running tasks in the background. According to a dev.to report on Google's announcement, the standard model is pitched as the cheaper, more scalable tier for live dialogue, while the Extended Thinking variant targets harder workloads that need deeper reasoning across multiple steps.

Reasoning without pausing the conversation

The defining feature of the lineup is that work can continue behind an active conversation. As dev.to describes it, Google says both models can reason in real time, take in visual context and execute background tasks without interrupting the dialogue.

That is a meaningful shift for voice interfaces. Most conversational AI today is effectively turn-based: the user speaks, the system stops to process, then responds. A model that can keep the thread of a discussion going while it checks information or completes a multi-step action opens up different interaction patterns. A customer-service assistant, for instance, could keep talking to a caller while retrieving account details, rather than leaving dead air or forcing a restart of the exchange.

Google also highlights multilingual language switching and tool calling as part of the broader live-dialogue toolkit, and cites integrations such as Docs Live, Gmail Live and Keep Live alongside its developer ecosystem, according to the report.

Availability, with gaps

dev.to lists several rollout routes for the new models:

• Developers can access them through the Gemini Live API and Google AI Studio.
• Gemini Enterprise customers get private previews, with Google noting that the models are headed to Gemini Enterprise for Customer Experience.
• General users can encounter them through Search Live and the Gemini app.

The report also flags what is not yet clear. Google has not published exact pricing or a full regional rollout schedule in the material provided, and the enterprise customer-experience deployment is explicitly described as forthcoming rather than shipped. Feature parity across every product surface should not be assumed, and teams evaluating implementation are advised to confirm availability, commercial terms and the current feature set for their chosen access route.

Audio provenance

Audio output from these AI products is watermarked with SynthID, Google's technology for helping detect AI-generated content, dev.to reports. For anyone publishing or distributing AI-generated audio, that offers a stated provenance mechanism, although it does not replace the need for clear internal rules on customer-facing communications.

Dreambeans widened the same week

Separately, Google Labs expanded Dreambeans, its daily personalized-story experiment, to all eligible Google accounts in the United States for users aged 18 and over, on Android and iOS. According to dev.to, the experience spans the Google AI Plus, Google AI Pro and Google AI Ultra access levels and can draw on Calendar, Gmail, Photos, Search, YouTube and now Gemini.

Unlike Gemini 3.8 Live, Dreambeans comes with no stated developer or enterprise workflow in the report. Its significance is consumer availability across the US and its growing set of Google-service integrations, which together show Google pushing AI interaction in two directions at once: more capable live conversation and more personalized consumer content.

Why it matters

The jump here is architectural as much as functional. Real-time voice AI has mostly meant fast question-and-answer; background reasoning turns voice into a plausible interface for tasks, not just queries. If the models deliver as described, conversational systems could handle lookups, tool calls and multi-step work inside a live dialogue, which is the difference between a talking FAQ and something closer to an agent with a voice. Google's broad rollout across APIs, enterprise previews and consumer apps signals that it wants this capability to spread quickly across its surfaces.

The caveats are real: pricing, regional timing and enterprise feature availability are still undefined, and the source notes that practical value will depend on how products wire these models into actual workflows, integrations and escalation paths rather than on voice alone. For developers and enterprises building voice products, Gemini 3.8 Live sets a new baseline for what live conversation AI is expected to do.