OpenAI Introduces Faster Realtime AI Models
OpenAI has released two new AI models, gpt-realtime-2.1 and gpt-realtime-2.1-mini, through its API. The company states the models will deliver at least a 25% reduction in p95 latency.
OpenAI has announced the release of two new models, gpt-realtime-2.1 and gpt-realtime-2.1-mini, available via its API. The company reports that these models offer a significant improvement in response times.
The primary focus of these models is on low-latency applications, particularly for voice agents and multimodal interactions. OpenAI states that through optimizations in their caching mechanisms, the p95 latency for the Realtime voice models has been reduced by at least 25%. This metric indicates the response time for 95% of requests.
The gpt-realtime-2.1-mini model supports real-time audio and text input, alongside tool usage and function calling. It is designed to perform internal reasoning before executing actions and maintain conversational continuity through verbal cues.
Additionally, the gpt-realtime-2.1 model features enhancements in alphanumeric recognition, handling of silence and noise, interruption management, and instruction following. Users can adjust the model's inference intensity across five levels, with a recommendation to start with lower settings for production environments.