OpenAI launches GPT-Realtime-2.1 with 25% lower voice delay

OpenAI has added two new low-latency voice models to its API: gpt-realtime-2.1 and gpt-realtime-2.1-mini. Official data shows that the p95 delay of these models has dropped by at least 25%.

Gpt-realtime-2.1-mini supports real-time audio and text input, with tool calling and function calling features. It can think before taking actions to keep conversations smooth. Gpt-realtime-2.1 improves letter and number recognition, silence and noise handling, and offers five levels of reasoning intensity.

For pricing, gpt-realtime-2.1-mini charges $0.60 per million tokens for text input and $10 for audio input. Gpt-realtime-2.1 costs $4 for text input and $32 for audio input. Both models aim to provide low-latency solutions for voice agents and multimodal interactions.