Talking to a customer-service bot is often frustrating: long pauses, lost context and a bot that cannot actually do anything. Companies building voice assistants want AI that can hold a natural conversation and complete the task at the same time.
That is the pitch behind Amazon Nova 2.5 Sonic. On 5 October 2026, Amazon Web Services (AWS) announced that the model, its latest speech-to-speech AI for real-time voice agents, is now generally available, meaning it is out of preview and open for production use [1].
What is Nova 2.5 Sonic?
A speech-to-speech model listens to spoken audio and replies in speech directly, rather than relying on a separate chain of tools to transcribe, think and then talk. A voice agent is an AI system that can hold a spoken conversation and take actions along the way.
AWS says the new version improves reasoning, instruction following and tool-calling accuracy, meaning how reliably it uses connected software to get things done, and also has lower latency, so responses come faster [1]. Its example: a customer-service agent that can look up an order, check whether it can be returned and start the return, all within one conversation.
What carries over from before?
AWS says Nova 2.5 Sonic keeps the existing capabilities of its predecessor [1]:
- expressive voices in seven languages;
- controllable turn-taking, which governs when the AI speaks and when it listens;
- voice and text in the same session;
- asynchronous tool calling, so it can keep talking while a task runs in the background;
- a 256K context window, the amount of conversation and information it can keep in view at once.
Developers can also use it with Strands Bidi Agents, which AWS says is now generally available too, to build production voice agents [1].
Where can you get it?
Nova 2.5 Sonic is available through Amazon Bedrock, AWS’s service for accessing AI models, in four regions: US East (N. Virginia), US West (Oregon), Europe (Stockholm) and Asia Pacific (Tokyo). AWS says pricing is the same as Nova 2 Sonic [1].
What we don’t know
- How much better it is. The improvements in reasoning, accuracy and speed are AWS’s own claims. The announcement gives no figures and no independent benchmark results.
- Wider availability. At launch it is limited to four regions, and AWS has not said when others will follow.
The Bottom Line
Nova 2.5 Sonic is an incremental but practical upgrade: AWS says its voice model is now quicker and better at actually getting tasks done, at no extra cost. Whether it feels noticeably better on a real customer call will depend on testing that AWS has not yet published.
Sources
- AWS What’s New, 5 October 2026: announcement of the general availability of Amazon Nova 2.5 Sonic. https://aws.amazon.com/about-aws/whats-new/2026/10/amazon-nova-2.5-sonic/
