Amazon's Nova 2.5 Sonic Voice Model Hits Bedrock in Four Regions
TL;DR
- Amazon Nova 2.5 Sonic went generally available on Bedrock on October 5, 2026, in US East, US West, Stockholm and Tokyo regions.
- The speech-to-speech model carries a 256K context window, offers expressive voices in seven languages, and allows voice and text in the same session.
- Pricing matches Nova 2 Sonic, and asynchronous tool calling lets voice agents trigger tasks like order lookup without stalling the conversation.
Amazon flipped Nova 2.5 Sonic to general availability on Bedrock on October 5, the latest in its speech-to-speech line aimed at real-time voice agents. The AWS announcement says the model carries a 256K context window, offers "expressive voices in seven languages," and supports voice and text in the same session.
Launch regions are US East (N. Virginia), US West (Oregon), Europe (Stockholm), and Asia Pacific (Tokyo). Pricing matches Nova 2 Sonic, though the post does not restate the figure. Asynchronous tool calling is the signature upgrade, with order lookup and return processing named as the kinds of tasks a voice agent can hand off without stalling the conversation.
AWS also flagged Strands Bidi Agents as now generally available, pitched as a way to "build production voice agents in a few lines of code." The post quotes no executive and publishes no benchmark numbers, so the "improved reasoning, instruction following, and tool-calling accuracy" claim lands as marketing copy rather than measured gain.
Originally reported by aws.amazon.com
Read the original article →Original headline: Amazon Ships Nova 2.5 Sonic Speech-to-Speech Model on Bedrock, 256K Context, Seven Languages, Async Tools