Amazon announced the general availability of Nova 2.5 Sonic on October 5, 2026. The model is a speech‑to‑speech engine designed for real‑time voice agents. It adds better reasoning, follows multi‑step instructions, and calls external tools more accurately. The new version also reduces the time between a user speaking and the agent responding, making conversations feel faster.
What the upgrade adds
Nova 2.5 Sonic can keep track of context across longer dialogs and handle tasks like checking an order, verifying return eligibility, and starting a return without leaving the conversation. It works in seven languages, supports both voice and text in the same session, and can handle up to 256,000 tokens of context. Developers can pair it with Strands Bidi Agents, which are also generally available, to build production voice agents with only a few lines of code.
Availability and pricing
The model is offered through Amazon Bedrock in four AWS regions: US East (N. Virginia), US West (Oregon), Europe (Stockholm), and Asia Pacific (Tokyo). Pricing is the same as the previous Nova 2 Sonic model.
Why it matters
For developers building voice assistants, the lower latency and stronger reasoning mean customers will experience faster, more natural conversations. Businesses can automate more complex tasks—like checking orders and processing returns—without human hand‑offs. Users will notice quicker responses but the exact cost impact depends on their usage patterns.
