Live Translation and Transcription Introduced By OpenAI Latest API Models

OpenAI redefines voice technology with a new generation of api models capable of real time reasoning translation and transcription for global developers

On May 7, OpenAI launched its new generation of voice models that are capable of reasoning, translating, and transcribing as users speak. Also, OpenAI’s latest API models potentially create a wide range of voice apps for developers. This means that developers can now create apps that can talk, transcribe, and translate conversations with users in real-time. The three new models are:

• GPT-Realtime-2: Comes with GPT-5 class reasoning and can handle harder quests along with carrying out natural conversations.

• GPT-Realtime-Translate: OpenAI’s new live translation model, which translates speech from over 70 input languages into around 13 output languages while maintaining the pace of the speaker.

• GPT-Realtime-Whisper: A new streaming speech-to-text that transcribes speech live as the speaker talks.

Furthermore, OpenAI noted that building practical voice products is a complex process, involving the AI agent understanding the context of conversations, adjusting to changing requests, using tools as conversations continue, and responding in an appropriate manner in different contexts. In terms of the target audience, organizations that seek to expand their customer service offerings stand to benefit the most by deploying these models. However, OpenAI emphasized its applicability in several other areas as well, including media, events, education, and creator platforms.

Moreover, real-time translation in the Indian market is expected to comprise services with multilingual voice experiences. Developers can use the model to build live multilingual voice experiences where individuals can speak in their desired language as well as hear the translated conversation and read transcripts in real time. According to Prateek Sachan, Co-founder and CTO at BolnaAI, “Building voice AI for India means handling diverse regional phonetics. In our evals across Hindi, Tamil, and Telugu, GPT-Realtime-Translate delivered 12.5% lower Word Error Rates than any other model we tested, along with lower fallback rates, higher task completion, and latency that sustained natural conversation.” Thus, CIO Bulletin views OpenAI’s three new models as fundamentally altering the paradigm of voice models and setting a new benchmark for multilingual voice AI.