OpenAI's new voice stack is an agent play, not a party trick

On 7 May, three models: GPT-Realtime-2 (first voice model with GPT-5-class reasoning), Translate (live, 70+ to 13 languages) and Whisper (streaming transcription). Translate and Whisper bill by the minute; Realtime-2 by tokens — a tell that OpenAI expects the reasoning one to do the heavy, expensive work.

Translation and transcription are increasingly commodity — plenty of labs do them well. The thing competitors can't trivially match is a voice model with frontier reasoning baked in, low-latency enough to hold a conversation. That's the piece that turns voice from 'dictation with a personality' into agents that can actually do tasks while you talk. OpenAI is selling the picks and shovels for the voice-agent gold rush it expects next.

Timing's no accident either: this dropped in early May, right before Sesame's voice app and Apple's Siri AI. Whoever owns the developer layer owns the ecosystem, and OpenAI moved first. The voices everyone coos over will be built on someone's API — OpenAI wants it to be theirs.