On Thursday, OpenAI announced that its Realtime API will introduce several voice intelligence features designed to help developers build applications capable of listening, speaking, translating, and transcribing. The core offerings include three new models. GPT-Realtime-2 leverages GPT-5-level reasoning capabilities to handle more complex user instructions and enable more natural voice conversations. GPT-Realtime-Translate supports real-time translation across over 70 input languages and 13 output languages, allowing instant switching during dialogue. GPT-Realtime-Whisper provides real-time speech-to-text functionality, completing transcription concurrently as conversation occurs. According to OpenAI, these new models are advancing real-time audio technology beyond simple "question-and-answer" interactions toward true voice interfaces that can listen, reason, translate, transcribe, and execute actions. Target application areas span customer service, education, media, events, and creator platforms among others. Addressing potential misuse risks, OpenAI stated that built-in guardrail mechanisms have been implemented; if harmful content guidelines are violated during a conversation, the system can automatically terminate it. Regarding pricing, both Translate and Whisper operate on a per-minute billing basis, while GPT-Realtime-2 charges based on token consumption—all integrated into the Realtime API.