Google launches Gemini 3.8 Live and 3.8 Live Extended Thinking voice models with real-time visual context and background task execution

Google DeepMind released two voice/audio models, Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking, rolling out from today across the Gemini API, Google AI Studio, Gemini Enterprise (private preview), Search Live, the Gemini Live app, and, for Extended Thinking, Workspace (Docs, Gmail, Keep) for AI Pro/Ultra subscribers. Claimed features include near real-time visual context, automatic switching across 97 languages mid-conversation, background tool execution without interrupting the conversation, and SynthID watermarking on audio output. Google cites third-party benchmark placements, including top spot on Artificial Analysis's Speech to Speech Quality Index (82.6) and a 97.7% score on Big Bench Audio. No pricing was disclosed despite repeated references to being "cost-effective."

This affects developers building voice agents (via API access), enterprises evaluating Workspace and Gemini Enterprise voice tools, and consumers using Search Live or the Gemini app. The benchmark numbers are independently checkable, which gives this more substance than a typical announcement, and if the SSQI and agentic task scores hold up under third-party testing, it strengthens Google's position against rivals like OpenAI's voice models and specialised players such as Sierra and LiveKit. But without pricing, nobody can yet judge whether it is commercially competitive, and the practical impact for most businesses will depend on how it performs outside controlled benchmarks.

The "most advanced" framing and repeated "cost-effective" claims are vendor packaging with no pricing to back them up. This is an iterative point release in a voice-model line that Google has been updating every few months, not a structural shift, so language implying a breakthrough moment should be discounted. The long list of partner names (Sierra, ServiceNow, Agora, LangChain, Vercel, etc.) reads as ecosystem-marketing padding rather than evidence of deep integration or committed usage.

1. 01Whether Google publishes concrete API pricing for Gemini 3.8 Live and Extended Thinking within the next few weeks, and how it compares to OpenAI's and Anthropic's voice offerings
2. 02Independent verification of the Artificial Analysis SSQI 82.6 and Big Bench Audio 97.7% scores by other benchmarking outlets or developers, not just Google's own citation
3. 03Whether any of the named partners (Sierra, ServiceNow, Salesforce, Genspark, Lumeris) publish their own case studies or usage data confirming real production deployment within the next one to two quarters