Unlocking the Power of Extended Thinking: Gemini Live Revolutionizes Gmail Experience

Gemini 3.8 Live Extended Thinking Features and Benchmarks

Gemini 3.8 Live Extended Thinking is built specifically for high-complexity tasks that demand increased intelligence and multi-step reasoning, according to Google’s documentation. The model powers features like conversational search in Gmail Live, draft generation and editing in Docs Live, and note creation in Keep Live. According to benchmark data provided in the company’s announcements, the extended thinking model captures the number one overall spot on Artificial Analysis’ Speech to Speech Quality Index with an 82.6 score. Furthermore, it leads in agentic task completion, achieving 68.6% on τ-Voice and 35.1% on Sierra’s τ-Voice-banking benchmark, while scoring 97.7% on Big Bench Audio.

Simultaneous Reasoning and Speech Capabilities

A core capability of the new extended thinking model is that it reasons and speaks simultaneously, according to official product details. This allows the system to deliver increased intelligence for complex workflows while maintaining an uninterrupted conversational flow. The model uses early verbal cues such as letting users know it is checking information to acknowledge prompts naturally. It also provides live progress narration to walk users through multi-step background tasks as they progress.

Gemini 3.8 Live Base Model and Multimodal Inputs

The base Gemini 3.8 Live model focuses on conversational intelligence featuring fluid dialogue and visual grounding, according to Google’s feature breakdowns. Based on DeepMind model cards, the underlying architecture is derived from Gemini 3 Pro and supports audio, images, video, and text inputs with a token context window of up to 128K, alongside a 64K token output. The model can process visual inputs in near real-time, executing tools and API calls in the background while conversation continues. Users can expect the model to acknowledge requests and keep chatting while tasks finish, alongside the ability to detect and transition between 97 supported languages mid-conversation. According to Google, the base model secures second place in the Speech Agent Arena and powers AI Mode’s Search Live experience while remaining cost-effective for scale.

Frequently Asked Questions

What are the primary differences between Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking?

According to Google, Gemini 3.8 Live is built for scale and cost efficiency with fluid dialogue and visual grounding, whereas Gemini 3.8 Live Extended Thinking is engineered for high-complexity workflows requiring multi-step reasoning, simultaneous speech, and background task narration.

What input modalities do the new Gemini 3.8 models support?

According to DeepMind model cards, both models accept audio, images, video, and text inputs with a context window of up to 128K tokens, producing audio and text outputs up to 64K tokens.

How does Gemini 3.8 Live Extended Thinking perform on industry benchmarks?

According to official metrics provided by Google, the extended thinking model ranks first on Artificial Analysis’ Speech to Speech Quality Index with an 82.6 score, leads τ-Voice agentic task completion at 68.6%, scores 35.1% on Sierra’s τ-Voice-banking benchmark, and achieves 97.7% on Big Bench Audio.