Technology
Google releases Gemini 3.8 Live dialogue models with enhanced reasoning
The new systems support multi-step reasoning, real-time voice processing across 97 languages, and audio watermarking.
The short version
- Google introduced two new conversational artificial intelligence models: Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking.[Hacker News · 9to5Google]
- Gemini 3.8 Live is designed for scale and can automatically switch between 97 languages mid-conversation.[Hacker News · 9to5Google]
- The Extended Thinking version targets complex multi-step reasoning and topped Artificial Analysis' Speech to Speech Quality Index.[Hacker News · 9to5Google]
- All audio output produced by the tools includes SynthID watermarks to assist in detecting synthetic media.[Hacker News]
Key facts
- Google officially introduced Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking on September 15, 2026.[Hacker News · 9to5Google]
- Gemini 3.8 Live Extended Thinking ranked first on the Speech to Speech Quality Index from Artificial Analysis with a score of 82.6.[Hacker News · 9to5Google]
- Gemini 3.8 Live Extended Thinking scored 68.6% on τ-Voice, 35.1% on Sierra's τ-Voice-banking benchmark, and 97.7% on Big Bench Audio.[Hacker News · 9to5Google]
- Gemini 3.8 Live can automatically detect and shift between 97 languages during an active dialogue and placed second in the Speech Agent Arena.[Hacker News · 9to5Google]
- All audio output generated by the models incorporates SynthID watermarking to help identify synthetic content.[Hacker News]
Sources
Outlet counts describe coverage, not independent confirmation. Reports may share a wire service or original source.