OpenAI launches GPT-4o, a faster and more affordable multimodal model

OpenAI has announced the launch of GPT-4o, its latest AI model designed to understand and generate text, audio, and images in an integrated manner. Known as "omni" for its multimodal capability, this model responds to audio inputs in just 232 milliseconds on average, comparable to human response time in conversations.

GPT-4o not only maintains the performance of GPT-4 Turbo in English and programming but also significantly improves performance in other languages, being faster and 50% cheaper in the API. Additionally, it shows notable advancements in visual and auditory comprehension compared to previous models.

Unlike previous models that used multiple models to handle audio, text, and vision inputs and outputs, GPT-4o uses a single neural network trained integrally to process all these types of data. This allows for more natural and rich interaction, capturing tones, multiple speakers, and background sounds, and can generate laughter, songs, and emotions.

The model has been evaluated and demonstrated superior performance in traditional text, reasoning, and coding benchmarks, as well as new visual and auditory perception tests. Additionally, OpenAI has implemented rigorous safety measures and risk assessments to mitigate potential dangers, ensuring that GPT-4o does not exceed medium risk in any safety category.

GPT-4o is immediately available to free users and ChatGPT Plus subscribers, with text and image capabilities. Soon, support for audio and video will be added to a select group of partners in the API. This launch marks a significant step towards more natural and efficient human-computer interaction.

For more information, visit the official announcement.