Updated
Updated · The Keyword | Google Product and Technology News · Sep 15
Google Rolls Out Gemini 3.8 Live Models Across 97 Languages as Extended Thinking Tops Speech Index at 82.6
Updated
Updated · The Keyword | Google Product and Technology News · Sep 15

Google Rolls Out Gemini 3.8 Live Models Across 97 Languages as Extended Thinking Tops Speech Index at 82.6

3 articles · Updated · The Keyword | Google Product and Technology News · Sep 15

Summary

  • Google began rolling out Gemini 3.8 Live and 3.8 Live Extended Thinking to developers, enterprises and users, positioning them as near-real-time voice AI models for more natural, production-ready agents.
  • 97 languages can be detected and switched between mid-conversation, while the models handle visual input in near real time and keep talking as they execute tools and API calls in the background.
  • 82.6 on Artificial Analysis' Speech to Speech Quality Index put Extended Thinking in the top overall spot; it also scored 68.6% on τ-Voice, 35.1% on Sierra's banking benchmark and 97.7% on Big Bench Audio.
  • Google said 3.8 Live is built for scale and cost efficiency, while Extended Thinking targets higher-complexity workflows in Search, Workspace and enterprise deployments including Docs, Gmail and Keep.
  • SynthID watermarking is being applied to all audio generated by Google's AI products, adding a detectability layer as the company expands voice features through its API, app and enterprise stack.

Insights

Can Google's imperceptible audio watermarks truly survive real-world tampering, or is SynthID just a regulatory illusion?
Will uninterrupted, multi-language voice AI finally replace human customer support, or simply generate faster hallucinations?
How does an AI that narrates its own thinking process manipulate our psychological trust in automated systems?