OpenAI adds GPT-Realtime-2, live translation, and streaming Whisper to its API

OpenAI voice API update

OpenAI has launched a new realtime voice stack for developers, adding GPT-Realtime-2 for harder spoken interactions, live translation across 70-plus input languages, and a low-latency streaming transcription model.

# OpenAI adds GPT-Realtime-2, live translation, and streaming Whisper to its API

## Opening summary

OpenAI is expanding its Realtime API with three new audio models that give developers a broader voice stack to build against. The release combines a stronger reasoning model for live spoken interactions, a translation model that keeps pace with a speaker, and a streaming transcription model meant to work as speech is happening.

## Main article

The headline addition is GPT-Realtime-2, which OpenAI says brings GPT-5-class reasoning to voice interactions. In the company’s own description, the model is meant to handle harder requests, recover more gracefully when something goes wrong, and keep a conversation moving while it calls tools or works through multi-step tasks.

OpenAI is also launching GPT-Realtime-Translate for live multilingual conversations and GPT-Realtime-Whisper for low-latency speech-to-text. According to the official post, the translation model supports more than 70 input languages and 13 output languages, while the transcription model is designed for live captions, notes, and other workflows that need words on screen immediately.

TechCrunch’s coverage helps sharpen the commercial angle. It notes that the update is aimed at developers building customer service, education, media, and event experiences, which makes this less about a flashy model release and more about OpenAI trying to become the default backend for production voice software.

## Why it matters

This matters because voice interfaces stop feeling gimmicky once they can reason through requests, translate across languages, and produce usable transcripts in the same session. OpenAI is clearly pitching that fuller stack to developers who want live voice products to do real work instead of just sounding smooth.

## Source notes

- Verified against OpenAI’s official product post, which names the three models, their roles, availability, and pricing basics. - Verified against TechCrunch, which confirms the release and its developer-facing positioning around transcribing, translating, and voice conversations. - The article keeps the scope to API availability and avoids implying a broad consumer feature launch.