Amazon Nova 2.5 Sonic

Amazon Nova 2.5 Sonic is AWS's latest native speech-to-speech model, reaching general availability in October 2026 through Amazon Bedrock. It delivers real-time bidirectional audio conversations with a 1 million token context window and adaptive prosody that mirrors the speaker's tone and cadence, with a median time to first audio of 1.08 seconds. The model covers 7 languages with polyglot voices that switch seamlessly across English in US, UK, India, and Australia accents, plus French, Italian, German, Spanish, Portuguese, and Hindi within a single session. Automatic language detection, robust background noise handling, graceful interruption recovery, and asynchronous tool calling allow agents to continue speaking while tool calls process in parallel. At $0.015 per minute, Nova 2.5 Sonic is the most cost-efficient native speech-to-speech model on AWS, with built-in connectors for Amazon Connect, Twilio, Vonage, and SIP trunking. It is the natural starting point for startup builders already within the AWS ecosystem who want low-cost, fully managed voice AI without assembling a cascade pipeline.