OpenAI today unveiled the GPT-4o model.

The most significant improvement lies in voice interaction. It now responds to conversations in real-time, with latency controlled at 232 milliseconds, nearing the average human reaction speed. The system can detect changes in tone and supports interruptions at any time.

In the demo video, the AI reads code with expressive intonation and immediately stops when the user interrupts. This level of fluency previously required specially developed voice assistants.

Text processing capabilities have also improved, though the official release did not provide specific data. It is currently available to free users, with paid users enjoying higher usage limits.

Interestingly, the model was not named GPT-5 as per convention. The team likely believes the core breakthrough lies not in parameter scale but in the qualitative leap in interaction experience.