OpenAI announced on Thursday that its API will now incorporate several advanced voice intelligence features, enabling developers to build applications capable of speaking, transcribing, and translating conversations with users in real time. The company's new model, GPT‑Realtime‑2, is a voice-focused tool designed to create lifelike vocal interactions. Unlike its predecessor, GPT-Realtime-1.5, this version is built with GPT‑5‑class reasoning, which OpenAI says allows it to handle more complex user requests. Additionally, OpenAI is launching GPT‑Realtime‑Translate, a feature that provides real-time translation services designed to keep pace with natural conversation flow. This tool supports over 70 input languages (the languages it can understand) and 13 output languages (the languages it can relay to the speaker). The company also introduced GPT-Realtime-Whisper, a new transcription capability that offers live speech-to-text functionality, capturing interactions as they happen.

EMBED_PLACEHOLDER_0

"Together, the models we are launching move real-time audio from simple call-and-response toward voice interfaces that can actually do work: listen, reason, translate, transcribe, and take action as a conversation unfolds," the company stated. So, who stands to benefit from these updates? Companies looking to enhance customer service capabilities are an obvious target. However, OpenAI also notes that these features can support a wide range of sectors, including education, media, events, and creator platforms, among others.

While these tools appear highly useful from an enterprise perspective, they also raise potential concerns about misuse. OpenAI says it has built in guardrails to prevent abuse, such as spam, fraud, or other forms of online harm. Specific triggers have been embedded in the system so that "conversations can be halted if they are detected as violating our harmful content guidelines," the company explained. All of the new voice models are available through OpenAI's Realtime API. Translate and Whisper are billed by the minute, while GPT-Realtime-2 is charged based on token consumption.