OpenAI launches three real-time speech models

OpenAI has launched three real-time speech models: GPT-Realtime-2, which features GPT-5 level inference and tool invocation capabilities; GPT-Realtime-Translate, which supports real-time translation between more than 70 languages at a cost of only about 0.25 yuan per minute, a hundred times lower than human simultaneous interpretation; and GPT-Realtime-Whisper, which achieves low-latency speech transcription. All three models are available through the Realtime API, providing end-to-end processing that preserves intonation and emotion.