OpenAI introduces three new audio-focused models designed for real-time voice applications. The announcements describe GPT-Realtime-2, aimed at real-time voice tasks, along with GPT-Realtime-Translate, which targets live speech translation. A third model, GPT-Realtime-Whisper, is presented as a real-time variant related to the Whisper speech-to-text approach. Across the reports, the common theme is that the models are built to handle audio inputs quickly enough for interactive use cases, rather than relying on slower, batch-style processing. The coverage does not indicate that these models replace existing OpenAI audio offerings, but instead positions them as additions for developers and product builders who want low-latency speech capabilities. Details beyond the model names and their intended purposes are not provided in the supplied excerpts, including specific performance benchmarks, supported languages, or deployment availability. The reports therefore focus primarily on the product lineup and what each model is designed to do: general real-time voice tasks, real-time translation, and real-time transcription.