This week, Google released two new AI models, Gemini 3.8 Live and Gemini 3.5 Transcribe, via the Gemini Live API. These models aim to help developers build low-latency conversational voice agents with multilingual support, visual context understanding, and more.
First, Gemini 3.8 Live is Google’s built-in model for fast verbal conversations. The model can run API and tool calls in the background while streaming audio responses without interruptions. It analyzes live images and videos at up to 1 frame per second to connect users’ speech with what they see.
This model supports 97 languages, with what Google claims are realistic accents and automatic mid-conversation language switching. Sessions last up to 15 minutes for audio-only and 2 minutes for audio and video. Pricing starts at $0.005 per minute for audio input and $0.018 per minute for audio output.
Gemini 3.5 Transcribe is Google’s dedicated speech-to-text model, designed for transcription. The model offers two API endpoints: the Live API for real-time streaming with sub-second-latency captions, and the Interactions API for pre-recorded audio files up to 1 hour, with diarization (the process of turning a transcript into clear segments attributed to individual speakers) and word timestamps.
Gemini 3.5 Transcribe can tackle 85 languages with automatic detection and regional accent handling. Google says its word error rate is around 4% for streaming and 2.6% for non-streaming.
Developers can access both models through Google AI Studio and the Gemini API. All AI-generated audio from Gemini Live models includes SynthID watermarking to help detect AI-generated content and prevent misinformation.
