Amazon Nova 2.5 Sonic: speech-to-speech model for voice agents (aws.amazon.com)

Amazon has launched the general availability of Amazon Nova 2.5 Sonic, an advanced speech-to-speech model designed to enhance the capabilities of voice agents in real-time applications. This latest iteration introduces improved reasoning, better instruction following, and heightened accuracy in tool usage, significantly reducing latency for more natural interactions. Notably, Nova 2.5 Sonic allows voice agents to maintain context during conversations and execute multi-step tasks, such as managing customer service queries seamlessly. The significance of Nova 2.5 Sonic lies in its ability to empower developers to create sophisticated voice agents across various domains, including customer service and personal AI assistants. Key features include enhanced speech understanding, multi-language support, controllable turn-taking, and a substantial 256K context window. It also integrates with Amazon's Strands Bidi Agents, simplifying the development of complex conversational applications with minimal code. Positioned within Amazon Bedrock and available across multiple AWS regions, Nova 2.5 Sonic offers an accessible pathway for the AI/ML community to elevate user interaction experiences through more capable and efficient voice technologies.