PolyAI has developed a new dialog model, Dialog-RSN-1, which directly processes caller audio instead of relying on ASR transcripts. This model integrates multiple functions, including turn-taking, speech recognition, and response generation, into a single audio-native system. The model also keeps text-to-speech separate to maintain control over output voice and operates as a request-based large language model. Initial tests have shown fast response times of under 300 milliseconds in live deployments. The development of this model is significant for improving the efficiency and effectiveness of conversational AI systems.