OpenAI adds GPT-5-level voice reasoning and real-time translation to its API

OpenAI announced Thursday that its API now includes a suite of new voice intelligence features, giving developers tools to build applications capable of natural conversation, live transcription, and real-time translation. The updates center on three new models — GPT-Realtime-2, GPT-Realtime-Translate, and GPT-Realtime-Whisper — each designed to handle different aspects of voice interaction.

GPT-Realtime-2 brings GPT-5 reasoning to voice

The flagship model, GPT-Realtime-2, succeeds GPT-Realtime-1.5 and is built on GPT-5-class reasoning. OpenAI says this enables the model to handle more complex user requests in real-time voice conversations, moving beyond simple call-and-response patterns. The company describes it as a realistic vocal simulation that can listen, reason, and respond contextually as a conversation unfolds.

Real-time translation across 70+ languages

GPT-Realtime-Translate offers conversational translation that keeps pace with natural speech. It supports more than 70 input languages — the languages it can understand — and 13 output languages for spoken responses. This positions the tool for use in international customer support, live events, education, and media localization, where speed and accuracy in spoken translation are critical.

Live transcription with Whisper

The third model, GPT-Realtime-Whisper, provides live speech-to-text capabilities that capture interactions as they happen. Unlike batch transcription services, this runs in real time, making it suitable for applications such as live captioning, meeting notes, and voice-controlled interfaces.

Enterprise applications and guardrails

OpenAI sees clear enterprise demand for these features, particularly in customer service automation. But the company also acknowledges misuse risks, including spam, fraud, and other forms of online abuse. To address this, OpenAI has embedded guardrails that can halt conversations if they violate harmful content guidelines. Specific triggers are built into the system to detect and stop abusive behavior.

Pricing and availability

All three models are available through OpenAI’s Realtime API. GPT-Realtime-Translate and GPT-Realtime-Whisper are billed by the minute of audio processed, while GPT-Realtime-2 is billed by token consumption, consistent with OpenAI’s existing pricing model for text-based models.

Why this matters

Voice interfaces have long been limited by latency and a lack of contextual understanding. OpenAI’s latest models aim to close that gap, making voice interactions feel more natural and capable of handling complex tasks. For developers, this means building apps that can transcribe, translate, reason, and act in real time — a step toward more human-like voice assistants. The updates also signal OpenAI’s continued push into multimodal AI, where voice, text, and reasoning converge in a single platform.

Conclusion

OpenAI’s new voice intelligence features represent a meaningful upgrade to its API, offering developers GPT-5-level reasoning, real-time translation, and live transcription in a single suite. With built-in guardrails and flexible pricing, the company is positioning these tools for broad enterprise adoption while addressing potential misuse. The updates are available now through the Realtime API.

FAQs

Q1: What is GPT-Realtime-2?
GPT-Realtime-2 is OpenAI’s latest voice model, built on GPT-5-class reasoning, designed for real-time, natural voice conversations that can handle complex user requests.

Q2: How many languages does GPT-Realtime-Translate support?
It supports over 70 input languages for understanding and 13 output languages for spoken responses.

Q3: How are the new voice models billed?
GPT-Realtime-Translate and GPT-Realtime-Whisper are billed by the minute, while GPT-Realtime-2 is billed by token consumption.