Announcing Amazon Nova 2.5 Sonic with improved reasoning for voice agents

Amazon Web Services announced the general availability of Amazon Nova 2.5 Sonic, a speech-to-speech model designed to enhance real-time voice agents with improved reasoning and instruction-following capabilities. The model reduces latency, enabling faster and more natural interactions during conversations. It supports maintaining context, executing multi-step instructions, and using connected tools to complete tasks such as order lookups and return processing within a single interaction.

Nova 2.5 Sonic builds on existing features including robust speech understanding, expressive voices in seven languages, and controllable turn-taking. It also supports voice and text within the same session, asynchronous tool calling, and a 256K context window for longer conversations. Developers can integrate these capabilities to create more capable voice agents for customer service, personal AI assistants, and real-time applications.

The model is available in Amazon Bedrock across four AWS Regions: US East (N. Virginia), US West (Oregon), Europe (Stockholm), and Asia Pacific (Tokyo). It is offered at the same pricing as Nova 2 Sonic, ensuring cost consistency for developers. This availability supports broader deployment of voice agents with consistent performance and pricing.

Developers can use Nova 2.5 Sonic alongside Strands Bidi Agents, now generally available, to build production voice agents with minimal code. The integration enables longer conversations and connections to other agents, streamlining the development process for real-time conversational applications.