VOICE AI

Google shipped two new audio-to-audio models on September 15: Gemini 3.8 Live (76.0 Speech-to-Speech Index, 1.18s first audio) and 3.8 Live Extended Thinking (82.6, #1), both with background tool execution. Benchmarks, per-minute pricing, and how to build with the Live API.

Voice cloning now requires just 3 seconds of audio to impersonate someone. With $893M in reported losses, detection has failed - here's what might actually work.

The Realtime API uses WebSockets for two-way voice interaction with function calling and stateful conversations. Here is how to set it up and build on it.