OpenAI released GPT-4o, a multimodal model capable of real-time voice conversation, image understanding and text generation, according to Reuters, setting a new standard for the responsiveness of consumer AI assistants.

The release matters less for any single capability than for what it made normal. A voice conversation with a model, with low latency and natural interruption, moved from a research demonstration to a shipped product, and every assistant that followed has been measured against the experience it established.

Why voice is the surface

Voice is the lowest-friction interface a human has, and a model that can hold a voice conversation removes the screen from a category of interactions that never needed one. The applications, from tutoring to customer service to companionship, are the ones the text-first assistants addressed awkwardly. The release did not solve those markets, but it gave the companies building for them the primitive they needed to start.