Google unveils Gemini 3.8 Live and 3.8 Live Thinking AI models

Google launched the Gemini 3.8 Live and Gemini 3.8 Live Thinking AI models, adding real-time audio transcription, reasoning and translation plus multi-step task handling, the company said on 17 September 2026. Google reported an average Word Error Rate of 4.0% for streaming transcription and 2.6% for non-streaming transcription, with support for 85+ languages and automatic detection of 97 languages mid-conversation. All audio generated by the new models is watermarked with SynthID to flag AI-generated content and curb misinformation.

Quick summary - click for full details Concise summary of key highlights In one line Google launches Gemini 3.8 Live and 3.8 Live Thinking AI models for real-time audio reasoning and multi-step tasks. Key points • Gemini 3 Enables real-time audio transcriptions, reasoning, and translations for multi-utility apps with near-instant visual input processing. • Gemini 3 Designed for high-complexity tasks with advanced multi-step reasoning and increased intelligence. • Gemini 3 Dedicated speech-to-text model offering highly precise transcription across 85+ languages with low error rates. • Voice command integration Users can invoke Gemini AI via smartphones to perform complex multi-step tasks across Google and third-party apps using voice commands. • Watermarking for AI content All audio generated by the new models is watermarked with SynthID to detect AI-generated content and prevent misinformation. Key statistics 4.0% Average Word Error Rate WER for streaming transcription 2.6% Average Word Error Rate WER for non-streaming transcription 85+ Number of languages supported by Gemini 3.5 Transcribe 97 Number of languages automatically detected mid-conversation Processed with AI. Reviewed by DH Digital Team. Published 17 September 2026, 05:28 IST