OpenAI releases GPT-4o Mini, a free version of its main AI model

OpenAI launched GPT-4o Mini this Thursday (18). The productive “mini model” The company’s most powerful AI is a scaled-down version of GPT-4o.

According to the company, this is “the most capable and efficient scaled-down model to date.” The model requires less energy to process data, making it a cheaper solution to implement in applications.

“I think GPT-4o Mini really fulfills OpenAI’s mission to make AI more accessible to people,” said Olivier Godement, OpenAI API Platform Lead. Border“If we want AI to benefit every corner of the world, every industry, every application, we must make the technology more accessible,” he continued.

The new model achieved an 82% score on the Measuring Massive Multitask Language Understanding (MMLU) assessment. The test consists of 16,000 multiple choice questions on 57 academic topics.

In the small model category, direct competitors like Google’s Gemini 1.5 Flash and Anthropic’s Claude 3 Haiku achieved 78.9% and 75.2% respectively. Gemini Ultra has the highest scoreWith a 90% success rate.

But, It is worth noting that the benchmark is only one of the comparison factors.but it is not an accurate representation of every AI’s performance.

Available on ChatGPT

HE The new GPT-4o Mini will be implemented on ChatGPT later today (18) It will be available to Enterprise users next week for free users and Plus and Team plan subscribers. The model will replace GPT-3.5 Turbo.

HE GPT-3.5 model will no longer be available in ChatGPThowever, developers can access it through OpenAI APIs, but Godement explained that the model will also be removed from the platform at some point.

Like its predecessor, GPT-4o Mini supports both text and images. OpenAI claims AI will soon handle multimodal inputs and outputsincluding video and audio.

However, unlike the entire model GPT-4o Mini is designed for simpler tasks. Therefore, it cannot meet the more complex needs for powerful virtual assistants, for example.