New Realtime models (GPT-realtime-2.1 and GPT-realtime-2.1-mini) on the API (community.openai.com)

A new suite of Realtime models has been released, featuring gpt-realtime-2.1 and its more compact version, gpt-realtime-2.1-mini. These models are designed to enhance low-latency voice and multimodal experiences, boasting a 25% reduction in p95 latency due to improved caching mechanisms. The gpt-realtime-2.1 model enhances its predecessor with better alphanumeric recognition, superior handling of silence and noise, and more effective interruption behavior. It supports advanced speech-to-speech interactions with customizable reasoning efforts, making it ideal for complex voice-agent workflows. In contrast, the gpt-realtime-2.1-mini provides a faster and cost-effective alternative for real-time voice applications. Users can choose the most suitable model based on their needs: for comprehensive reasoning and tool usage opt for gpt-realtime-2.1, while those prioritizing speed and affordability may prefer gpt-realtime-2.1-mini. These developments signify a notable advancement in the AI/ML community, enabling developers to create more sophisticated and efficient voice interaction systems, with a focus on performance and user experience.