Google released new audio models for developers. Gemini 3.8 Live enables speech-to-speech tasks while maintaining dialogue. 3.8 Live Extended Thinking adds deeper reasoning for complex requests. Both rank first on Artificial Analysis’ Speech-to-Speech leaderboard. Gemini 3.5 Transcribe offers precise transcription in 85+ languages. It achieved a 4.0% Word Error Rate for streaming. The models support asynchronous function calling and visual context. They also handle multilingual input and alphanumeric precision. Pricing starts at $0.005 per minute for audio input. Developers can access them via the Live API or Google AI Studio. Integration partners include Agora, Fishjam, LiveKit, LangChain, Pipecat, Vercel and Vision Agents.
What changed
Previous live models required cascaded architectures for complex tasks.
What it unlocks
Building voice agents that reason, execute tasks, and maintain conversation flow in real time.
- $0.005/min for audio input
- $0.018/min for audio output
- 4.0% WER streaming
- 2.6% WER non-streaming
What you need to act on it
- access to the Live API or Google AI Studio
Sources