Our organization relies heavily on GPT-Realtime in production, and after extensive testing, we have not found a single real-time model that can replace it.

We’ve evaluated Realtime 1.5, 2, and 2.1, and all of them consistently exhibit noticeable pronunciation and enunciation issues that make conversations sound less natural and less professional. In contrast, GPT-Realtime remains the only model that delivers consistently clear, accurate, and natural speech.

Deprecating GPT-Realtime before a true replacement exists puts businesses like ours in a difficult position. If the newer models still can’t match the voice quality, pronunciation accuracy, and overall conversational experience of GPT-Realtime, then they are not yet a functional replacement.

Please don’t deprecate GPT-Realtime until there is a model that demonstrably matches or exceeds its real-world voice performance. For organizations building production voice applications, speech quality is not a nice-to-have, it’s a core requirement.