Nova Sonic was Amazon's entry into speech-to-speech modelling: instead of chaining a transcriber, a language model and a synthesiser, it takes audio in and returns audio out in one bidirectional stream, keeping tone, pacing and interruptions intact. It went live on Amazon Bedrock in spring 2025 and became the standard building block for AWS voice agents in call centres and customer service. In December 2025 Amazon replaced it with Nova 2 Sonic and moved the original to legacy status: new customers can no longer adopt it and AWS switched it off on 14 September 2026 — the model was removed from every region and requests to it now fail.
#speech to speech#voice agents#real-time#AWS Bedrock#retired