Frontier Models
Google Gemini 3.8 Live Advances Real-Time Voice AI with Extended Thinking
The September 15 2026 release introduces models optimized for intuitive voice agents, background task handling and multi-language fluidity while rolling out across Google platforms with SynthID protections.
Gemini 3.8 Live is Google's live dialogue model built for scale and cost efficiency that combines conversational intelligence with fluid dialogue and visual grounding.
Google released Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking on September 15 2026.
The models target voice agents with advancements in near real-time reasoning.
Conversations become more intuitive and intelligent through these updates.
What core features support real-time voice interactions in Gemini 3.8 Live?
Gemini 3.8 Live supports near real-time visual input processing.
The model performs automatic mid-conversation transitions across 97 languages.
Asynchronous background tool and API calls run without interrupting dialogue.
These elements allow fluid collaboration on tasks using voice commands.
How does Gemini 3.8 Live Extended Thinking handle high-complexity tasks?
Gemini 3.8 Live Extended Thinking targets high-complexity tasks with increased intelligence.
Multi-step reasoning capabilities expand while maintaining uninterrupted conversational flow.
Parallel reasoning upgrades make the model more intuitive for complex task execution.
Users gain the ability to execute intricate operations through voice without breaks.
| Variant | Primary Focus | Key Capability | Benchmark Highlight |
|---|---|---|---|
| Gemini 3.8 Live | Conversational scale and efficiency | Near real-time visual input and 97-language transitions | Fluid dialogue with visual grounding |
| Gemini 3.8 Live Extended Thinking | High-complexity reasoning | Multi-step and parallel reasoning | 82.6 on Speech to Speech Quality Index and 68.6 percent on τ-Voice |
| Gemini 3.8 Flash | Reasoning and coding | Cost-efficient performance | Introductory pricing of 0.75 dollars input and 3.75 dollars output per million tokens |
Which benchmarks show leadership for the Extended Thinking model?
The same model leads agentic task completion on the τ-Voice benchmark at 68.6 percent.
These results reflect gains in intelligence and parallel reasoning.
What rollout platforms receive the new Gemini 3.8 models first?
- Gemini API integration for developer access.
- Google AI Studio availability for testing and prototyping.
- Search Live and Gemini Live product integrations.
- Gemini Enterprise private preview for enterprise users.
- Google Workspace rollout for subscribers.
Gemini 3.8 Live and Gemini 3.8 Live Extended are major upgrades in intelligence and parallel reasoning that make them more intuitive to collaborate with and use to execute complex tasks using your voice.Tom Ouyang and Malini Jaganathan, Principal Engineer and Member of Technical Staff (on behalf of the Gemini Audio Team), Google
How does SynthID integration affect audio outputs from the models?
All audio generated by the models receives SynthID watermarks.
The watermarking enables detectability of AI-generated speech.
This measure supports transparency requirements in deployed voice systems.
What pricing structure applies to the companion Gemini 3.8 Flash release?
Gemini 3.8 Flash carries introductory pricing of 0.75 dollars per million input tokens.
Output tokens are set at 3.75 dollars per million through the end of 2026.
These performance gains stem from a core design choice: 3.8 Flash works harder.Tulsee Doshi and Raluca Ada Popa, Senior Director, Product Management and Gemini Security Lead, Google DeepMind
What market implications arise for developers and enterprises from the release?
Developers gain access through the Gemini API for building voice agents.
Enterprises receive private preview access via Gemini Enterprise.
Google Workspace subscribers obtain the models for internal productivity tools.
The rapid release cadence of three Flash versions in six weeks signals ongoing iteration.
What future steps are anticipated for the Gemini audio lineup?
The models add to the Gemini 3 series of natively multimodal reasoning systems.
Further refinements in real-time capabilities are expected in subsequent updates.
Background task handling and language support may expand based on current foundations.
Frequently asked
When were Gemini 3.8 Live and Extended Thinking released?
Google released Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking on September 15 2026 as part of the frontier model wave.
Which platforms receive the initial rollout of the new models?
The models roll out to the Gemini API, Google AI Studio, Search Live, Gemini Live, Gemini Enterprise private preview and Google Workspace for subscribers.
What benchmark scores does Gemini 3.8 Live Extended Thinking achieve?
Gemini 3.8 Live Extended Thinking scores 82.6 on the Speech to Speech Quality Index and leads with 68.6 percent on the τ-Voice agentic task completion benchmark.
Sources
- Google — Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking released on September 15 2026 with near real-time reasoning advancements for voice agents.
- Google — Gemini 3.8 Flash offers introductory pricing of 0.75 dollars per million input tokens and 3.75 dollars per million output tokens through end of 2026.
- Google DeepMind — Gemini 3.8 Audio models including Live and Live Extended Thinking form part of the Gemini 3 series of natively multimodal reasoning models.