OpenAI has released a new AI language model. It replaces GPT-3.5, released at the end of 2022, in all versions of ChatGPT.

As its name suggests, GPT-4o mini is a derivative of GPT-4o, OpenAI’s most powerful model. Introduced in May, GPT-4o (“o” for “omni”) can reason across audio, vision and text in real time.

Until now, it was available to all, including in the free version of ChatGPT, but on a limited basis. Now, in the free version of ChatGPT, GPT-4o mini becomes the standard model. In the various paid versions, users will have the choice between using GPT 4, GPT 4o and GPT-4o mini.

In developing GPT-4o mini, OpenAI says it took great care to filter out certain types of information, such as hate speech, adult content and spam. The model already supports text and vision, although its knowledge base currently ends in October 2023.

According to OpenAI, GPT-4o mini excels particularly in mathematical reasoning and coding tasks. Support for text, image, video and audio inputs and outputs will be coming in the future.

For developers, this is by far the most cost-effective model, as GPT-4o mini costs just 15 cents per million input tokens (the fundamental units of text used by large language models) and 60 cents per million output tokens – the equivalent of 2,500 pages in a standard book.