INTRODUCING GPT-4o MINI: AN ADVANCEMENT TO SMALLER, FASTER AND COST EFFICIENT INTELLIGENCE.
We have big, certainly, wonderful news from Open AI.
On July 18, Open AI introduced a new GPT model — the GPT-4o Mini. A powerful, breathtaking model, that has gotten almost everyone in the AI community talking.
Some people are happy about the new prospects, saying that they believe that this is what humans need from AI- a smaller, efficient, model. But, as there are good commentaries, there are also negative ones. Are all these new developments worth the hype? Can the newest version of the GPT-4o live up to the expectations of the masses? All these and many other questions bother some people, inasmuch as some others are simply happy on Open AI’s new developments.
Now, I know most people are going to be like, “Well, we don’t need newer versions of GPT-4, we need GPT-5. I understand your point. But let’s look at the bigger picture. What we need is not a higher model. What we need is a more EFFICIENT model. A CHEAPER model. And most importantly, a FASTER model. This is what the GPT-4o Mini embodies. What makes the GPT-4o Mini even better is that it has a higher reasoning task( both text and vision) than the rest of the smaller models, making an impressive score of 82.0% on MMLU, a textual intelligence and reasoning benchmark, as compared to 77.9% for Gemini Flash and 73.8% for Claude Haiku.
GPT-4o mini excels in mathematical reasoning and coding tasks, outperforming previous small models on the market. On MGSM, measuring math reasoning, GPT-4o mini scored 87.0%, compared to 75.5% for Gemini Flash and 71.7% for Claude Haiku.
To show you how cost efficient this is, 15 cents input/ 60 cents per output per 1 million tokens is huge, and this unlocks a lot of potential applications that would have been too expensive to operate well before. Also, for us to have things like agents and real time voices (upcoming), we can’t do it at the cost of the current cutting edge models- they’re still far too expensive.
But the million dollar question is; will GPT-4o Mini stand the test of time? Is it just a grand scheme by Open AI to attract more users all over the world to help “promote” the application. Is GPT-4o Mini really as good as they say it is?
Well, it’s safe to say that GPT-4o Mini is still a young, promising app, with lots of surprises coming out way. We in the AI community are as curious as anybody and are waiting for how things turn out in the following weeks. In my opinion, I can proudly say that GPT-4o Mini has brought quite a lot of opportunities that will benefit the masses and it does look promising indeed. But what if I’m wrong? Will GPT-4o Mini really live up to the standard Open AI has created for it?
Written by Sonia Ifeancho
Source : https://openai.com/index/gpt-4o-mini-advancing-cost-efficient-intelligence/