OpenAI releases GPT-4, a multimodal AI that the manufacturer claims is the most advanced.

OpenAI has released a powerful new image and text understanding AI model, GPT-4, which the company calls "the latest milestone in its efforts to scale deep learning."

GPT-4 is available today to paid OpenAI users through ChatGPT Plus (with usage restrictions), and developers can sign up for a waiting list to access the API.

The price is $0.03 for 1,000 "hint" tokens (about 750 words) and $0.06 for 1,000 "completion" tokens (again, about 750 words). Tokens represent the raw text; for example, the word "fiction" will be split into "fan", "tas", and "tic" tokens. Hint tokens are the parts of words submitted to GPT-4, while completion tokens are the content created by GPT-4.

According to the company, OpenAI spent six months "iteratively fine-tuning" GPT-4 using lessons from its internal competitive testing program as well as ChatGPT, resulting in "the best results" in terms of factuality, manageability, and refusal to go outside the fence. As with previous GPT models, GPT-4 was trained using publicly available data, including from public web pages, as well as data licensed by OpenAI.

OpenAI worked with Microsoft to develop a "supercomputer" from scratch in the Azure cloud that was used to train GPT-4.

"In casual conversation, the difference between GPT-3.5 and GPT-4 can be subtle," OpenAI wrote in a blog post announcing GPT-4. "The difference comes when the complexity of the task reaches a sufficient threshold - GPT-4 is more robust, creative, and capable of handling much finer instructions than GPT-3.5."

Without a doubt, one of the most interesting aspects of GPT-4 is its ability to understand images as well as text. GPT-4 can sign - and even interpret - relatively complex images, such as identifying a Lightning Cable adapter from an image of a connected iPhone.

However, OpenAI notes that it has made improvements in certain areas; GPT-4 is less likely to reject queries about how to synthesize dangerous chemicals, for example. The company claims that GPT-4 is 82% less likely to respond to requests for "prohibited" content compared to GPT-3.5 and responds to sensitive requests - such as medical advice and anything related to self-harm - in accordance with OpenAI's policy 29% more often.