OpenAI introduced a lightweight, powerful AI model GPT-4o Mini – it will replace GPT-3.5 for all users

OpenAI has officially unveiled a simplified and more accessible version of its most powerful large language model, GPT-4o. We are talking about the GPT-4o Mini neural network, the use of which will be more profitable for developers from an economic point of view, but at the same time its performance is higher than that of GPT-3.5.

Image source: Growtika / unsplash.com

Building applications using OpenAI language models can be a huge expense. In the absence of sufficient funds, developers may abandon their use in favor of more affordable analogues, such as Gemini 1.5 Flash from Google or Claude 3 Haiku from Anthropic. Now with the launch of GPT-4o Mini, the competition in this segment will be more intense.

«I think GPT-4o Mini really fits with OpenAI’s mission to make AI more accessible to people. If we want AI to bring value to every corner of the world, every industry, every application, we must make AI much more accessible,” said an OpenAI representative.

Starting today, ChatGPT Free, Plus, and Team users can use GPT-4o Mini instead of GPT-3.5 Turbo, and Enterprise users will have the option next week. This means that the GPT-3.5 language model will no longer be available to ChatGPT users, but will still be available to developers via the appropriate API if they do not plan to migrate to GPT-4o Mini. It is still unknown when exactly access to GPT-3.5 will be closed completely.

The GPT-4o Mini language model supports text, images, audio and video content. At the same time, the algorithm is still designed to solve simple problems, so it can be useful for adding various narrowly focused functions to third-party applications. In the MMLU benchmark, OpenAI’s new language model scored 82%. For comparison, the result of GPT-3.5 in the same test is 70%, GPT-4o – 88.7%, Gemini Ultra – 90%. The language models competing with GPT-4o Mini, Claude 3 Haiku and Gemini 1.5 Flash, scored 75.2% and 78.9%, respectively, in this test.