PolyAI Dialog-RSN-1 Arrives on AWS Marketplace and SageMaker JumpStart: One-Click Real-Time Voice Conversation Model, English Only for Now

AWS announced that PolyAI's real-time conversational reasoning model, Dialog-RSN-1, is now available on AWS Marketplace and SageMaker JumpStart and can be deployed in a customer's own AWS account. English only for now; connections last up to 30 minutes.

News date: About 5 min read

Share on LINE
Contents
  1. 1. What Is Dialog-RSN-1, and How Does It Differ from Traditional Voice Agents?
  2. 2. How Do You Get It? Which Deployment Options and Specs Are Supported?
  3. 3. How Does It Affect Existing Workflows?
  4. 4. FAQ

AWS announced on its Machine Learning Blog that PolyAI's Dialog-RSN-1 is now available on AWS Marketplace and Amazon SageMaker JumpStart. JumpStart is the machine learning model hub inside SageMaker AI. AWS calls Dialog-RSN-1 PolyAI's first model on AWS Marketplace. AWS says it can be used starting today and mentions no preview or waitlist, but it does not state which regions are covered. For teams building phone-based voice agents, this means they can deploy a real-time conversational model using their existing AWS billing, networking and permission setup.

What Is Dialog-RSN-1, and How Does It Differ from Traditional Voice Agents?

AWS describes Dialog-RSN-1 as a reasoning model built for complex real-time conversations: it can listen to a call, reason about it and act in real time. AWS says it perceives the caller's audio, reasons about their needs and responds in a single flow. There is no separate speech-to-text step and no separate model deciding when to speak, which reduces the number of handoffs.

AWS says that because the model processes audio directly, it can capture signals that a plain transcript would lose, such as tone, hesitation and timing, and it decides for itself when to respond. AWS also says barge-in still works during peak call volume. Barge-in means the caller can interrupt while the system is speaking. AWS adds that transcripts come from the audio itself. These are AWS and PolyAI product descriptions. The AWS post gives no performance figures to compare against.

How Do You Get It? Which Deployment Options and Specs Are Supported?

AWS describes two paths: - Subscribe through AWS Marketplace and deploy the model as a SageMaker endpoint. An endpoint is an online service that lets applications connect to and call the model. - Deploy directly from SageMaker JumpStart. In the SageMaker AI console, open Model Hub, select Dialog-RSN-1 from the JumpStart catalog, then choose Deploy. Automated workflows can also use the SageMaker Python SDK, and AWS provides a GitHub repository that explains the code path.

Key specifications of Dialog-RSN-1 on SageMaker (compiled from the AWS blog)
ItemAs described by AWS
AvailabilitySubscribe via AWS Marketplace; deploy via SageMaker JumpStart / Studio
ProtocolOpenAI Realtime protocol, through PolyAI's bridge by changing the base URL
Audio input16-bit PCM (uncompressed digital audio) mono, 16 or 24 kHz; G.711 or Opus must be transcoded first
Connection lengthUp to 30 minutes
LanguageEnglish only for now
Optimization profilesLatency-optimized, throughput-optimized, cost-optimized, balanced
Governance and monitoringYour own VPC, IAM roles and CloudWatch

How Does It Affect Existing Workflows?

AWS says users still own orchestration, telephony integration and business logic, while the model handles understanding the live call and responding. The endpoint runs in the customer's VPC (virtual private cloud, the customer's isolated private network on AWS). Access is controlled with IAM (Identity and Access Management) roles, and the endpoint uses the customer's existing CloudWatch logs and metrics. Charges are billed through the customer's AWS account. AWS says inference runs inside the customer's account, so caller audio and transcripts stay within a boundary the customer controls. AWS says this helps support data residency and audit requirements.

For evaluation, AWS says the JumpStart catalog provides an overview and sample notebooks. After deploying the endpoint, you can run PolyAI's sample notebook, which streams recorded call audio and outputs the model's turn-taking decisions, replies and transcripts. Before deployment, JumpStart also shows three metrics to help you choose an ml.* instance size: - P50 latency: the time within which half of requests complete. - Time to first token: how long the model takes to produce the first piece (token) of its output. - Throughput.

FAQ

Do I need to prepare my own speech-to-text service for Dialog-RSN-1?

As AWS describes it, the model receives the caller's audio directly, with no separate speech-to-text step, and the transcript is produced from the audio itself. Telephony integration and business logic remain the user's responsibility.

Can existing Realtime clients use it directly?

AWS says the endpoint supports the OpenAI Realtime protocol, so existing Realtime clients can connect through PolyAI's bridge after changing the base URL.

How do JumpStart's four optimization profiles differ?

Latency-optimized targets interactive real-time calls and reduces time to first token. Throughput-optimized raises tokens per second and suits high call volume. Cost-optimized lowers cost per request. Balanced is a general-purpose compromise.

Does it support languages other than English?

AWS says the model currently supports only English and does not mention a timeline for other languages.

How much does it cost?

The AWS blog gives no pricing; it only says charges are billed through the customer's AWS account.

Browse the latest news in this topic

Found this guide helpful? Share it with someone who needs it

Share on LINE

Sources

Compiled from the references and verification dates listed above; fares, schedules, product features, and terms can change, so please confirm the latest details on the relevant official website before travelling, booking, or purchasing.

Lifestyle