GPT Audio is OpenAI's multimodal audio model designed for native speech-to-speech interaction via the Chat Completions API. Unlike traditional voice pipelines that chain separate speech-to-text and text-to-speech models, GPT Audio processes and generates audio directly through a single model, resulting in lower latency, more natural-sounding voices, and better preservation of speech nuances such as tone and emotion.