OpenAI launches small AI model GPT-4o Mini. What is it and why is it important?

Sam Altman led AI startup OpenAI has announced GPT-4o, the company’s most cost effective model to date. The model is being seen as an attempt by OpenAI to stay relevant with increasing competition from more deep pocketed rivals like Google and Meta.

GPT-4o Mini (O stands for Omni), will replace GPT-3.5 Turbo and will be available to use starting today for free along with ChatGPT Plus and Team members. Meanwhile, it will be offered to enterprise users starting next week.

Also Read | Anthropic rivals GPT-4o with Claude 3.5 Sonnet model, makes it free for all users

OpenAI said that GPT-4o Mini is priced at 15 cents per million input token and 60 cents per million output tokens, making it 60% more cheaper than GPT-3.5 Turbo. The model scored 82% on Massive Multitask Language Understanding (MMLU) and outperformed GPT-4 on chat preferences in LMSYS leaderboard.

The company also claimed that GPT-4o Mini also comprehensively defeated other small models in reasoning tasks with Gemini Flash only managing a MMLU score of 77.9% and Claude Haiku a score of 73.8%.

Announcing the new model in a blog post, OpenAI wrote, “OpenAI is committed to making intelligence as broadly accessible as possible. Today, we’re announcing GPT-4o mini, our most cost-efficient small model. We expect GPT-4o mini will significantly expand the range of applications built with AI by making intelligence much more affordable.”

Also Read | SIMBA, the new AI-powered tool adopted by Nagpur police for crime detection. All you need to know

Small models like GPT-4o require low computational power and hence are a more affordable option for devleopers with limited resources who want to use generative AI in their applications.

GPT-4o Mini currently support text and vision in application programming interface (API) and support for text, image, video and audio outputs will be made available in the future, OpenAI said.

The latest model has a context window of 128K token, which translates to around 95,000 words, and has a cut off date of October 2023. Meanwhile, OpenAI stated that GPT-4o Mini is even more cost effective handling non-English text now owing to the improved tokenizer.