Google announced two new audio-capable Gemini models designed to improve near‑real‑time reasoning and make spoken interactions with AI more natural: Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking. Both models are positioned as production-ready building blocks for developers and enterprises building voice-driven agents, and as a way to make conversations with Gemini across the Gemini app, Google Workspace, and Search more fluid and collaborative.
Model distinctions and capabilities
- Gemini 3.8 Live: optimized for scale and cost efficiency, combines conversational intelligence with smooth dialogue flow and visual grounding.
- Gemini 3.8 Live Extended Thinking: aimed at high‑complexity tasks, offering enhanced multi‑step reasoning and higher task intelligence.
Gemini 3.8 Live can process visual inputs in near real time to enrich conversational context, and it supports automatic mid-conversation detection and switching among 97 supported languages. The model can execute tools and API calls in the background while continuing the conversation, allowing it to acknowledge requests and keep chatting while tasks finish asynchronously.
3.8 Live Extended Thinking is designed to reason while speaking: it provides increased intelligence for complex workflows without interrupting the conversational flow, using early verbal cues (for example, “Let me check that…”) and live progress narration to inform users about multi‑step background tasks.
Benchmarks and performance
Google reports that Gemini 3.8 Live Extended Thinking achieved top results on several metrics: it ranked first overall on Artificial Analysis’ Speech to Speech Quality Index with a score of 82.6. On agentic task completion measures it scored 68.6% on the τ‑Voice benchmark and 35.1% on Sierra’s τ‑Voice‑banking benchmark. For reasoning, the model scored 97.7% on Big Bench Audio. Google emphasizes these outcomes come with a competitive price point relative to other frontier models.
Gemini 3.8 Live also placed highly in user preference testing, taking second place in the Speech Agent Arena, while remaining cost‑effective according to Google.
On ServiceNow’s EVA‑Bench (an evaluation benchmark for voice agents), Google says these models push the Pareto frontier for complex workflows by balancing accuracy and conversational quality. The company ran these evaluations on the Live API of the Gemini Enterprise Agent Platform.
Demonstrations and use cases
Google highlights several demo scenarios: real‑time employee onboarding that leverages visual context, near‑real‑time chess play using visual inputs and reasoning, transforming raw sketches and live voice feedback into working React components with 3.8 Live Extended Thinking, coordinating multi‑step bookings and asynchronous function calls without interrupting conversation, and generating complete business plans and custom marketing toolkits on the fly via natural speech.
The Live models are integrated with Google Workspace and Search surfaces; 3.8 Live Extended Thinking is available in Docs Live, Gmail Live, and Keep Live in supported configurations, which Google positions as especially helpful for complex tasks.
Developer and enterprise ecosystem
Using the Gemini Live API, developer platforms such as Agora, Fishjam, LangChain, LiveKit, Pipecat, Vercel, and Vision Agents can enable developers to build and deploy high‑performance voice interfaces. These platforms handle the complex real‑time media streaming infrastructure so developers can focus on user experience.
Google also notes partnerships with companies including Salesforce, Genspark, and Lumeris, which have highlighted the models’ low latency, fluidity, and tool‑calling capabilities.
Safety and provenance
All audio generated by Google’s AI products will include an imperceptible SynthID watermark embedded in the audio output to help ensure AI‑generated content can be detected and to mitigate misinformation. Google refers readers to the model card for details on safety and responsible use.
Availability and rollout
Google states these releases begin today:
- Gemini 3.8 Live: available to developers via the Gemini API and Google AI Studio; enterprise customers in private preview in Gemini Enterprise and coming soon to Gemini Enterprise for Customer Experience; and for general users via Search Live.
- Gemini 3.8 Live Extended Thinking: available to developers via the Gemini API and Google AI Studio; enterprise customers in private preview in Gemini Enterprise and coming soon to Gemini Enterprise for Customer Experience and Google Workspace business customers; and available to consumers on Gemini Live and for Google AI Pro and Ultra subscribers in Workspace (Docs), and for all Google AI subscribers in Gmail and Keep.
Google indicates the models are immediately available to developers while enterprise and consumer availability will roll out across products and subscription tiers.
