OpenAI launches new voice intelligence features in its API

OpenAI announced on Thursday an expansion of its API with several new voice intelligence features, equipping developers to build applications with more sophisticated conversational capabilities. The release is aimed at moving real-time audio interfaces beyond simple commands toward interactive systems that can reason, translate, and transcribe during a live conversation. This update signals a significant push to enable a new class of voice-driven applications that can perform complex tasks for users.

New Real-Time Voice Capabilities

The new models are accessible through OpenAI’s Realtime API and provide a suite of tools for developers. The core additions include a more powerful conversational model and specialized tools for translation and transcription.

• GPT-Realtime-2: A new voice model for creating realistic vocal simulations. Unlike its predecessor, it integrates GPT-5-class reasoning to handle more complicated user requests and conversational flows.
• GPT-Realtime-Translate: Provides real-time translation that keeps pace with a live conversation, comprehending over 70 input languages and speaking in 13 output languages.
• GPT-Realtime-Whisper: Offers live speech-to-text transcription, capturing interactions as they occur.

The pricing structure for these new features varies. Both Translate and Whisper are billed by the minute of audio processed, whereas the more advanced GPT-Realtime-2 model is billed based on token consumption.

These updates primarily target enterprise use cases, with customer service being an obvious application. However, OpenAI also noted their relevance for sectors like education, media, events, and creator platforms. Acknowledging the potential for misuse, the company stated it has built in guardrails, including triggers designed to halt conversations if they violate harmful content guidelines, to mitigate risks such as spam, fraud, or other forms of abuse.

By embedding GPT-5-class reasoning directly into its real-time voice model, OpenAI is moving beyond basic transcription and translation services. This positions its API as a foundational layer for developers building sophisticated, task-oriented voice agents that can handle complex conversational workflows, directly challenging specialized enterprise AI solutions.