Overview
GPT-Realtime-Translate is a streaming speech-to-speech translation model for live multilingual audio experiences. OpenAI announced it on May 7, 2026, for use through the Realtime API. (openai.com)
Capabilities and uses
- Accepts spoken audio, automatically detects the source language, and streams translated audio and transcript updates while the speaker is still talking. Developers specify the target language. (developers.openai.com)
- The launch announcement states support for more than 70 input languages and 13 output languages. The documented target languages are Spanish, Portuguese, French, Japanese, Russian, Chinese, German, Korean, Hindi, Indonesian, Vietnamese, Italian, and English. (openai.com)
- Intended uses include multilingual calls, customer support, cross-border sales, education, events, broadcasts, and media. The dedicated endpoint is
v1/realtime/translations. (openai.com)
Limitations
- Text is output-only; image and video inputs are not supported. Function calling, structured outputs, fine-tuning, and predicted outputs are not supported. The documented context window is 16,000 tokens, with a maximum of 2,000 output tokens. (developers.openai.com)
- Audio already in the selected target language may not be translated. A production application may need original-audio passthrough or mixing to avoid silence. Two-way conversations generally require a separate translation session for each direction. (developers.openai.com)
Pricing
OpenAI lists USD $0.034 per minute of realtime audio duration for live translation, rather than text-token billing. Its pricing table does not list separate input and output token prices for this model; the published duration price should not be interpreted as separate input and output charges. Billing granularity beyond the stated per-minute rate was not verified. Pricing checked November 16, 2026. (developers.openai.com)