Voice agents have been expensive to deploy and cumbersome to manage because models can’t handle conversation, and because context ceilings force enterprises to build layers of session resets, state compression, and reconstruction at every deployment. OpenAI’s three new voice models are designed to reduce this burden, and they’re changing how engineers can think about translating voice into a larger stack of agents.
GPT-Realtime-2, GPT-Realtime-Translate, and GPT-Realtime-Whisper integrate real-time audio into the model management stack as discrete orchestration primitives—separating speech reasoning, translation, and transcription into separate components instead of combining them into a single audio product.
The company said blog post Realtime-2 is the first voice model with “GPT-5 class reasoning” and can handle difficult queries and carry on conversations naturally. Realtime-Translate understands more than 70 languages and translates them into another 13 languages at the speed of a speaker, and Realtime-Whisper is its new speech-to-text transcription model.
These three actions no longer sit within a single stack or model. GPT-Realtime-2 can technically handle transcription, but OpenAI offloads different tasks to specialized models: Realtime-Translate for multilingual speech and Realtime-Whisper for transcription. Rather than routing everything through a single, all-encompassing voice system, businesses can assign each task to an appropriate model.
New OpenAI models compete Voxtral models of Mistralwhich separates the transcription and target enterprise use cases.
What businesses should do
Now that more people are comfortable conversing with an AI agent, and also because of the wealth of data gained from voice customer interactions, more businesses are seeing the value of voice agents.
Organizations evaluating these models should consider not only model quality, but also their orchestration architecture, specifically, can their stack route discrete voice tasks to specialized models and manage state in a 128K-token context window?
