Skip to content
trendoviz

PolyAI Dialog-RSN-1 launches on SageMaker JumpStart

PolyAI's Dialog-RSN-1, a reasoning model for real-time, complex dialogue, is available through AWS Marketplace and Amazon SageMaker JumpStart. Customers can deploy it as an inference endpoint in their own AWS account, using their existing VPC, IAM roles, and CloudWatch monitoring.

51 out of 100Emerging7D: Insufficient history ProvisionalPublished

Overview

PolyAI Dialog-RSN-1 is now offered through AWS Marketplace and Amazon SageMaker JumpStart, with deployment also available directly from SageMaker Studio [1]. It is the first PolyAI model offered through AWS Marketplace [1].

The model is positioned for live customer-call handling: PolyAI describes it as perceiving caller audio, reasoning about what the caller needs, and responding in a single pass, with no separate speech-to-text step or multiple handoffs [1]. This description is a vendor_claim [1]. PolyAI also states that the model is already handling live calls in production, though the announcement gives no named customers, deployment volumes, dates, or independent verification for that claim [1].

Why it matters

The deployment model moves the operational question from hosting a speech model to managing a marketplace subscription and a SageMaker endpoint inside an existing AWS account [1]. Inference happens within the customer's account, and AWS states that caller audio and transcripts stay within a boundary the customer controls, which the announcement links to data residency and auditability requirements [1]. This is a vendor_claim; no audit results are provided [1].

PolyAI says customers retain ownership of orchestration, telephony integration, and business logic, while the model handles live-call understanding and response [1]. That separation of ownership is a vendor_claim, but it means teams keep control of the surrounding contact-center stack rather than handing the entire call path to a single vendor [1].

Opportunities

Teams with an existing OpenAI Realtime client may be able to reduce integration work by pointing that client at PolyAI's bridge through a base-URL change rather than rebuilding the client [1]. The integration still depends on PolyAI's bridge, whose implementation is not described in the announcement [1].

The sample notebook and catalog materials create a path for evaluating the model against a team's own recorded calls before committing to it [1]. For teams building voice agents, PolyAI claims the model detects tone, hesitation, and turn timing directly from caller audio, signals a plain transcript would lose; whether that holds for a given call flow is an open question for the deploying team [1].

The named optimization profiles offer a starting point for tuning a real-time serving workload, though the specific profiles are not enumerated in the announcement [1].

Risks

Pricing, subscription terms, usage charges, and infrastructure costs are not provided, so the total cost of operating a live-call endpoint cannot be assessed from the announcement [1].

The input constraints are firm: 16-bit PCM mono audio at 16 or 24 kHz, a 30-minute connection ceiling, and English-only operation at announcement time [1]. These limit multilingual deployments, calls that require G.711 or Opus decoding upstream, and long-duration use cases [1].

The announcement does not specify model size, architecture, training methodology, context limits, regional availability, service quotas, or hardware requirements, which makes capacity planning and migration planning difficult [1]. Adopting a third-party model through AWS Marketplace also introduces dependency on both PolyAI and AWS for continued availability and support [1].

Context

Dialog-RSN-1 is packaged for teams that already run contact or telephony systems on AWS. Subscription happens through AWS Marketplace, and the model runs as a SageMaker endpoint inside the customer's AWS account rather than as a separate PolyAI-hosted service [1].

The endpoint speaks the OpenAI Realtime protocol, so an existing Realtime client can connect through PolyAI's bridge by changing its base URL [1]. Input audio must be 16-bit PCM mono at 16 or 24 kHz, with G.711 or Opus transcoded first; connections last up to 30 minutes, and the model was English-only at announcement time [1].

The JumpStart catalog entry includes an overview, sample notebooks, and named optimization profiles for workload tuning [1]. A PolyAI sample notebook streams recorded caller audio and prints turn decisions, replies, and transcripts [1]. The announcement says JumpStart deployment is a one-click process that creates a real-time inference endpoint on a preconfigured, already-tuned serving stack, with the same path available through a few lines of the SageMaker Python SDK; this is a vendor_claim [1].

Known, claimed, inferred

The practical change is the distribution path. Dialog-RSN-1 is now packaged as a managed cloud endpoint, so evaluation and operational controls remain with the deploying team even though the model and serving layer come from PolyAI [1]. AWS describes the serving stack as preconfigured and tuned, with named optimization profiles, but the profiles and configurations are not enumerated in the available text [1].

A reasonable hypothesis is that the immediate decision for an AWS-based contact team is operational rather than architectural: whether a marketplace subscription and SageMaker endpoint fit existing deployment, cost, and governance processes [1]. The absence of published measurements leaves that judgment open, because the announcement supports no comparison between this deployment model and alternatives for handling live calls [1].

The protocol compatibility claim is the clearest lever on integration cost, since base-URL substitution is a smaller change than a new client implementation, though the bridge remains an undocumented dependency in the available material [1].

Limitations & uncertainties

There are no benchmark scores, accuracy figures, latency measurements, or comparisons with other dialogue models, and no independent or third-party evaluation results are included [1]. PolyAI's statement that Dialog-RSN-1 is already handling live calls in production is not supported by named customers, deployment volumes, dates, or independent verification [1].

The identity of PolyAI's bridge, the details of the OpenAI Realtime protocol integration, and the implementation of the one-click deployment are not further specified [1]. Regional availability, service quotas, hardware requirements, model size, architecture, training methodology, and context limits are not stated [1]. The available material comes from a single AWS announcement, and the second source is a shorter version of the same post rather than an independent report [1].

Evidence & sources

The key statements of this analysis, labeled by type, with the sources they rely on. Trendoviz does not run product tests.

  • Primary-source or vendor statementPolyAI Dialog-RSN-1 is available through AWS Marketplace and Amazon SageMaker JumpStart, and is the first PolyAI model offered through AWS Marketplace.[1]
  • Primary-source or vendor statementCustomers can deploy the model as a SageMaker endpoint within their own AWS account, using their VPC, IAM roles, and CloudWatch logging and metrics.[1]
  • Primary-source or vendor statementThe endpoint speaks the OpenAI Realtime protocol, allowing an existing Realtime client to connect through PolyAI's bridge by changing its base URL.[1]
  • Primary-source or vendor statementAudio input is 16-bit PCM mono at 16 or 24 kHz, connections last up to 30 minutes, and the model was English-only at announcement time.[1]
  • Primary-source or vendor statementPolyAI describes Dialog-RSN-1 as reasoning directly over caller audio in a single pass, without a separate speech-to-text step or multiple handoffs.[1]
  • Primary-source or vendor statementPolyAI states that the model is already handling live calls in production and detects tone, hesitation, and timing.[1]
  • Primary-source or vendor statementAWS describes JumpStart deployment as a one-click process that creates a real-time inference endpoint on a preconfigured, already-tuned serving stack, also available through a few lines of the SageMaker Python SDK.[1]
  • Primary-source or vendor statementThe JumpStart catalog entry includes an overview, sample notebooks, and named optimization profiles, and a PolyAI sample notebook streams recorded caller audio while printing turn decisions, replies, and transcripts.[1]

Sources

  1. 1
    PolyAI Dialog-RSN-1 is now on Amazon SageMaker JumpStart
    ●aws.amazon.comretrieved Nov 19, 2026

Updates & corrections

  1. First published · published automatically after passing every check

Found an error? Report it.