Google rolls out Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking models

What the New Models Offer

Google announced Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking as its most advanced live‑dialogue models to date.

The two models are built to make voice interactions feel more natural, fluid, and intelligent.

Gemini 3.8 Live focuses on fast, fluid conversations with real‑time visual and language support.

Gemini 3.8 Live Extended Thinking adds higher‑order reasoning, allowing the system to manage complex, multi‑step tasks while the user continues speaking.

Both models can process real‑time visual context and execute background tasks without breaking the flow of conversation.

According to the launch announcement, “We are launching Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking to make voice interactions more natural, fluid, and intelligent.”

The Gemini Audio Team, represented by Principal Engineer Tom Ouyang and Member of Technical Staff Malini Jaganathan, highlighted the models’ ability to handle interruptions and language switches.

Developers can access the models today via the Gemini API, Google Workspace, and the Gemini app.

Performance Benchmarks and Cost

Gemini 3.8 Live Extended Thinking achieved the top spot on Artificial Analysis’ Speech‑to‑Speech Quality Index with a score of 82.6.

The model also led the τ‑Voice benchmark with a 68.6 % success rate and posted 35.1 % on Sierra’s τ‑Voice‑banking test.

In the Big Bench Audio evaluation, Gemini 3.8 Live Extended Thinking scored 97.7 % for reasoning capability.

Gemini 3.8 Live placed second in the Speech Agent Arena, indicating strong user preference.

Both models are positioned as cost‑efficient, with Gemini 3.8 Live described as “built for scale and cost efficiency.”

On ServiceNow’s EVA‑Bench, the models pushed the Pareto frontier for complex workflows by balancing accuracy with conversational quality.

Despite the high performance, Google maintains a competitive price point relative to other frontier models.

Availability and Use Cases

Enterprises can leverage the models to build production‑ready voice agents that execute tasks such as scheduling, data retrieval, and multi‑tool orchestration.

Within Google Workspace, users can ask Gemini to draft documents, generate presentations, or pull information from spreadsheets using only voice.

In Google Search, the models enable more interactive, conversational answers that incorporate visual elements when relevant.

The Gemini app now supports continuous dialogue where the AI explains its thought process while working on a request.

Developers can integrate the models into custom applications through the Gemini API, extending voice‑first experiences to new domains.

By handling background tool usage without interrupting the user, the models aim to reduce friction in voice‑centric workflows.

Google frames the launch as a step toward AI that “listens and thinks along with you,” positioning the technology as a practical conversational partner.

Early adopters are encouraged to experiment with the models to assess suitability for specific voice‑assistant scenarios.

Overall, the release reflects Google’s broader strategy to embed advanced generative AI across its consumer and enterprise product portfolio.

Why This Matters: Google’s Gemini 3.8 Live models enable more natural, real‑time voice interactions and complex task handling, expanding practical AI use in enterprise and consumer products.