Google Releases Gemini 3.8 Live and 3.5 Transcribe to Enhance Real-Time Voice AI Performance

Google has released Gemini 3.8 Live and Gemini 3.5 Transcribe to support the development of real-time, voice-first applications. Gemini 3.8 Live represents a step change in native speech-to-speech models, enabling complex task execution while maintaining conversational flow.

Key Features of Gemini 3.8 Live

Gemini 3.8 Live Extended Thinking provides deep reasoning for complex requests and has ranked first on Artificial Analysis's Speech-to-Speech leaderboard. Key features include Asynchronous function calling, which maintains audio streaming during background API calls, real-time visual input-based conversations, and multilingual support for over 97 languages and accents. It also offers Alphanumeric precision for parsing confirmation codes and Incremental content updates that merge real-time audio with structured data.

Performance and Features of Gemini 3.5 Transcribe

The dedicated speech-to-text model Gemini 3.5 Transcribe supports over 85 languages, achieving high accuracy with an average WER of 4.0% in streaming mode and 2.6% in non-streaming mode. It is optimized for low-latency, high-precision tasks such as call center agents and real-time audio analysis through Automatic code-switching, which handles language switching within sentences, and Custom vocabulary biasing, which corrects up to 1,000 terms. Via the Interactions API, it also supports transcription of audio files up to one hour long and speaker diarization.

Deployment Infrastructure and Pricing

Gemini 3.8 Live audio input is priced at $0.005 per minute, and audio output at $0.018 per minute. It integrates with major deployment infrastructure partners such as Agora, LiveKit, and Vercel.