> ## Content Index > Fetch the complete content index at: https://eazzytechnews.ghost.io/llms.txt > Use this file to discover other available public pages before exploring further. # Gemini 3.8 Live keeps voice agents working while you talk - URL: https://eazzytechnews.ghost.io/google-gemini-38-live-voice-agents/ - Published: 2026-09-21T20:16:10.000Z - Updated: 2026-09-25T13:08:45.000Z - Description: Google says Gemini 3.8 Live and 3.8 Live Extended Thinking make real-time voice agents more useful by combining visual grounding, background tool calls and deeper reasoning. - Author: Collins Anfo - Tags: Google, Gemini, Voice AI, AI Agents, Developer Tools Google says Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking can keep a spoken conversation moving while the model uses visual context, reasons through harder requests and runs tool calls in the background. The [Google Blog announcement](https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-8-live-gemini-3-8-live-extended-thinking/?ref=eazzytechnews.ghost.io), dated September 15 and updated September 17, frames the two models as its most advanced live dialogue systems. The company says Gemini 3.8 Live is built for scaled voice interactions, while Gemini 3.8 Live Extended Thinking is built for higher-complexity tasks that need multi-step reasoning. The practical change is that voice agents can acknowledge a request, keep talking and continue work behind the scenes. That matters for customer support, workplace assistants and developer-built voice products because a user may ask for a booking, an inbox action, a troubleshooting step or a visual explanation without waiting in silence while every tool call finishes. ## What Google says changed Google describes three capabilities that move Gemini Live closer to a working agent rather than a voice-only chatbot. | Capability | What it means for users | Source limit | | --------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------ | -------------------------------------------------------------------------------------------------------------------- | | Visual context | The model can use live visual input during a spoken exchange, so a person can point at a screen, object or document and ask about what they see. | Google describes the capability; the article does not provide independent accuracy tests for every setting. | | Background tool calls | The model can start API or tool work while continuing to stream audio, reducing the dead time common in multi-step voice flows. | Google gives examples and partner claims, but real production reliability will depend on each integration. | | Extended Thinking | The higher-reasoning model can handle more complex workflows and narrate progress while it works through steps. | Google cites benchmark positions and scores; those remain benchmark evidence, not proof of every deployment outcome. | The [developer post](https://blog.google/innovation-and-ai/technology/developers-tools/build-real-time-voice-applications-gemini-audio/?ref=eazzytechnews.ghost.io) adds more implementation detail. It says the models support asynchronous function calling, visual grounding, alphanumeric precision, multilingual support across 97-plus languages and incremental content updates. It also says Gemini 3.8 Live Extended Thinking supports configurable thinking for multi-step reasoning in the background. ## Availability is split by audience For developers, Google says both Gemini 3.8 Live models are available through the Gemini API and Google AI Studio. The developer post says they are available through the Live API, with estimated audio pricing of $0.005 per minute for input and $0.018 per minute for output, based on token pricing footnoted by Google. For enterprises, Google says the models are in private preview in Gemini Enterprise and are coming to Gemini Enterprise for Customer Experience. For consumer-facing use, Google says Gemini 3.8 Live is rolling out in Search Live, while Gemini 3.8 Live Extended Thinking is rolling out in Gemini Live and to Google AI subscribers in parts of Workspace such as Docs, Gmail and Keep. Those availability labels matter. "Rolling out" and "private preview" do not mean every user or enterprise tenant has the same access on day one. They mean Google has started distribution across several product surfaces while keeping some business use cases behind preview channels. ## Why background work is the agent test A real-time voice system has a stricter user-experience problem than text chat. If the agent needs to call a booking API, search a file, inspect a camera feed or update a record, the person on the other end can feel the delay immediately. Google is trying to make the model respond naturally while that work continues. The useful mechanism is simple: speech stays active while tool work happens elsewhere. Google gives examples such as onboarding help, chess with visual context, React component creation from sketches and multi-step bookings with asynchronous function calls. Those are demonstrations of possible workflows, not a guarantee that every external system will integrate cleanly. Google also says all audio generated by its AI products is watermarked with SynthID. That is relevant because more natural speech output can make provenance harder for listeners to judge. Watermarking does not solve every misuse problem, but it gives Google a technical control it can point to as generated audio becomes easier to deploy. ## What to watch next The strongest next evidence will come from deployed voice products: latency under load, failed tool-call handling, language-switching quality, escalation to a human and user consent around camera or screen context. Enterprise customers will also care whether background tasks leave a clear audit trail when an agent changes a record, sends a message or books something on behalf of a user. For now, Google's confirmed development is narrower and still consequential. Gemini's live voice line is being pushed from quick spoken answers toward agents that can see context, talk through uncertainty and keep working while the conversation continues. **Author:** [Collins Anfo, a digital product builder. My write-ups cover technology and artificial intelligence, from software, intelligent agents and robotics to chips, computing infrastructure, cybersecurity, consumer devices and space technology. I also cover technology-driven discoveries across mathematics, science and other research fields, product releases and announcements, and the business, finance, investment, governance and real-world adoption of technology.](https://collins-anfo-portfolio-2026.collinsanfo24.chatgpt.site/?ref=eazzytechnews.ghost.io) **AI assistance disclosure:** AI tools assisted with the research and drafting of this article. Material claims are linked to their sources for independent verification.