PANews reported on July 19 that according to Titanium Media, in the early morning of July 19, Beijing time, the US OpenAI company launched a new AI model "GPT-4o mini", which is a smaller parameter and simplified version of GPT-4o. OpenAI said that GPT-4o mini is currently the most powerful and cost-effective small parameter model, with performance close to the original GPT-4, and the cost is more than 60% cheaper than GPT-3.5 Turbo. It supports 50 different languages and replaces the GPT-3.5 version of ChatGPT, but the company did not disclose its training parameter scale information. OpenAI said that at the API level, GPT-4o mini supports 128k and 16k input tokens (images and text); at the price level, GPT-4o mini is 15 cents (about 1.09 yuan) per million input tokens and 60 cents (about 4.36 yuan) per million output tokens.
According to the plan, GPT-4o mini is now open to ChatGPT free users and ChatGPT Plus and Team paid subscribers, and will be open to ChatGPT Enterprise customers next week. In addition, OpenAI will also open a fine-tuned version of GPT-4o mini in the next few days. OpenAI CEO Sam Altman also revealed that just two years ago, the best model in the world was the text-davinci-003 version of GPT-3, which performed much worse than today and was 100 times more expensive. In other words, in just two years, the cost price per token of the GPT large model has dropped by 99%.
