Amazon has made Nova 2.5 Sonic generally available, with AWS touting better reasoning, instruction-following and tool-calling for real-time voice agents. Developers can use it in four AWS Regions, at the same price as Nova 2 Sonic, and pair it with Strands Bidi Agents, now generally available.
AWS AI Watch analysis
What happened
Nova 2.5 Sonic is a speech-to-speech model for voice applications. AWS says it also reduces latency and can maintain context, follow multi-step instructions and use connected tools during a conversation. Its example is a customer-service agent that looks up an order, checks return eligibility and starts a return without breaking the conversation.
The model supports voice and text in the same session, asynchronous tool-calling and a 256,000-token context window. AWS says it offers expressive voices in seven languages and controllable turn-taking. It is available in the US East (N. Virginia), US West (Oregon), Europe (Stockholm) and Asia Pacific (Tokyo) Regions, at the same price as Nova 2 Sonic. AWS says developers can use it with Strands Bidi Agents to build production voice agents.
Why it matters
The promise here is not simply a voice that sounds more natural. A voice agent that can keep track of a conversation, follow several steps and use tools could complete a customer task rather than merely explain where to click. Lower latency matters too: a clever voice assistant that leaves long pauses is still a rather awkward assistant.
For developers already building on Amazon Bedrock, the release adds a more capable option without a price increase over Nova 2 Sonic, according to AWS. The announcement describes the improvements, but does not give independent benchmark results or quantify the latency change.
Our read
This is a meaningful update for teams building voice agents, especially where the agent needs to act on information rather than just talk about it. The useful test is whether the claimed gains make multi-step tasks more reliable in real conversations, not whether a demo can order its own lunch. Developers can try the model in the four supported Regions and compare it with their current voice workflow.
What to watch
- Whether developers find the reasoning and tool-calling improvements hold up on real customer tasks.
- How the claimed latency reduction compares with Nova 2 Sonic in practice.
- Whether AWS expands regional availability or publishes more performance detail.
Discussion spark: For a voice agent handling customer tasks, which matters most: reliable tool use, natural conversation, or low latency?
Sources and evidence
- Announcing Amazon Nova 2.5 Sonic with improved reasoning for voice agents (5 October 2026, 08:10 UTC)
not affiliated with or endorsed by Amazon Web Services (AWS)