Google unveiled Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking on September 15, 2026 – two new voice AI models designed to enable more natural and intelligent speech interactions. The models can handle complex tasks in the background without interrupting conversations, setting a new standard for voice-powered AI agents.
The essentials
- Gemini 3.8 Live Extended Thinking leads Artificial Analysis' Speech-to-Speech Quality Index with 82.6 points
- Model achieves 68.6% on τ-Voice and 35.1% on Sierra's τ-Voice-banking benchmark for agentic task completion
- Available immediately via Gemini API, Google Workspace, and the Gemini app
- Led by Tom Ouyang (Principal Engineer) and Malini Jaganathan (Technical Staff) of the Gemini Audio Team
Two models for different use cases
Google differentiates between two variants: Gemini 3.8 Live is designed for scalability and cost efficiency, combining conversational intelligence with fluid dialogue and visual grounding. Gemini 3.8 Live Extended Thinking targets complex tasks with increased intelligence and multi-step reasoning.
Both models can handle interruptions, switch between languages, and explain their reasoning process while working. This approach aims to make AI interaction more intuitive and less frustrating for users.
Benchmark leadership and performance
Gemini 3.8 Live Extended Thinking delivers strong benchmark results: the model scores 97.7% on Big Bench Audio and leads multiple industry benchmarks. These numbers suggest Google is currently at the forefront of voice AI quality – particularly significant when compared to OpenAI's GPT-Live-1, which is positioned as a competing product.
| Metric | Score |
|---|---|
| Speech-to-Speech Quality Index | 82.6 |
| τ-Voice (Agentic Task Completion) | 68.6% |
| Sierra's τ-Voice-banking | 35.1% |
| Big Bench Audio | 97.7% |
Availability and access
The new models are available immediately for developers and enterprises – via the Gemini API for custom integrations, Google Workspace for business users, and the Gemini app for individual users. This means organizations can deploy production-ready voice interfaces today without lengthy development cycles.
What this means for enterprises
For organizations globally, voice AI is becoming a commodity feature. Anyone planning customer support, internal assistants, or voice commerce solutions can now leverage proven, production-ready models without waiting for proprietary development. Gemini 3.8 Live Extended Thinking's benchmark leadership makes Google the preferred choice for raw voice quality. However, real-world performance on domain-specific tasks and industry-specific use cases remains to be tested in the coming weeks.
Sources
Editorially owned by Ideal Syka. Sources and method: Newsroom & method. Tips and corrections: ai@i6eal.de.
