Nova 2 Sonic is the second generation of Amazon's speech-to-speech model, launched on 2 December 2025 and now the default choice for voice agents built on Amazon Bedrock. It keeps the single-stream design of its predecessor — audio in, audio out, no transcription hop in the middle — and adds Portuguese and Hindi to the original five languages. Its distinguishing feature is polyglot voices: one voice identity, for example Tiffany, speaks every supported language with native expressivity and can move between them inside a single conversation, so a caller switching languages no longer switches agents. The model carries a one-million-token context window and up to 64,000 output tokens, which is what allows a voice agent to hold a long call together.
#speech to speech#voice agents#real-time#AWS Bedrock#multilingual