Google Releases Gemini 3.8 Live Voice Models With Real-Time Reasoning
Google is now shipping a voice model that holds a conversation and completes coding tasks at the same time, and the third-party benchmark result puts it ahead of the rival voice models named in the same comparison.
Reporting from 1 source: GIGAZINE.
Google released two voice conversation models, Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking. Gemini 3.8 Live targets cost efficiency and Gemini 3.8 Live Extended Thinking adds multi-step reasoning. Both support 97 languages, including Japanese, converse in real time, and are available through API. In tests by the third-party organization Artificial Analysis, Gemini 3.8 Live Extended Thinking scored above OpenAI's GPT-Live-1-Astra and SpaceXAI's Grok Voice Think Fast 2.0.
Google put both models behind its Live API for developers. The demo videos show the split the company is selling: one model keeps a spoken conversation going while following a standing instruction such as responding only when addressed, and the other codes while the user talks to it.
The evaluation graph plots task completion rate against conversation quality, and Google positions the pair as reaching both at once rather than trading one for the other. Artificial Analysis ran the comparison against GPT-Live-1-Astra and Grok Voice Think Fast 2.0.
Synthesized by Yomimono from the 1 cited source below, including Japanese-language reporting where cited, then editorially reviewed before publishing.