OpenAI announced a series of model and product updates on 13 May 2024, introducing GPT-4o, a new flagship model the company describes as faster and more capable across text, vision, and audio [source].

The update brings native multimodal processing to the model, which OpenAI states can handle voice, text, and image inputs in a single conversation. Response times for audio inputs are reported as low as 232 milliseconds, comparable to human conversation speed. The model is being rolled out to ChatGPT Plus and Team users immediately, with availability for free-tier users following in phases.

GPT-4o replaces GPT-4 Turbo as the default model in the API. OpenAI states the new model is twice as fast and 50 percent cheaper than its predecessor, with improved performance on non-English languages and vision tasks. The company also announced a desktop application for macOS and updates to the ChatGPT interface.

The announcement includes a new voice mode that allows real-time spoken conversation with the model, though this feature is being released to a limited group of Plus users initially. OpenAI states the voice mode can respond to interruptions and detect emotional tone.

The update follows reports of performance variability in earlier GPT-4 releases, which users documented through systematic testing over the past year. OpenAI has not published detailed benchmarks comparing GPT-4o behaviour to previous versions under identical prompting conditions.

The company states GPT-4o will be made available to Enterprise users and through the API in coming weeks, with free-tier access subject to usage limits.