OpenAI’s new o3 and o3-mini models show progress towards “Artificial General Intelligence”

OpenAI has presented its new family of Artificial Intelligence (AI) models, o3 y o3-mini, that offer a improved performance with respect to previous models and are closer to the “Artificial General Intelligence” (AGI).

The company led by Sam Altman has unveiled the successor models to the o1 family within the framework of its latest announcement within the 12 days of OpenAI what they plan launch “at the end of January” in the case of o3-mini model and later in the case of o3.

Specifically, like the o1 models, the new models of the o3 family continue to expand the reasoning abilitiesdedicating more time to think before responding to improve your skills in solving complex tasks and problems. Thus, when a request is launched, or 3 analyzes related indications and explain your reasoning before offering a definitive answer.

Day 12: Early evals for OpenAI o3 (yes, we skipped a number)https://t.co/iWXg9IGuZM

— OpenAI (@OpenAI) December 20, 2024

Likewise, these new AI models offer performance improvements compared to its predecessorsFor example, based on American Invitational Mathematics Examination test results, o3 achieves an accuracy score of 96.7 percent, compared to 83.3 percent who get o1as the company has highlighted in an explanatory video on YouTube.

As OpenAI’s senior vice president of research, Mark Chen, explained, these performance improvements are equivalent to o3 “often fails on a single question”, which demonstrates more precision in its responses than previous models.

The same happens in the rest of the usual tests that OpenAI subjects its models to, in which o3 also obtained good results, so the company resorted to new, more challenging tests to put it to the test.

For example, the technology company used the ARC-AGI testspecially developed to measure the capabilities of AI models in comparison to the human mind, with which he managed to achieve a score of 87.5 percenta high figure considering that the 85 percent threshold is comparable to human performance.

In this sense, OpenAI has detailed that o3 represents an advance for Generative Artificial Intelligence, and that, even with his lowest score on the ARC-AGI test, he achieved triple the performance of o1.

MODEL O3-MINI

On the other hand, OpenAI has also presented o3-mini. Specifically, it is based on a smaller and more refined version of the modeloptimized to carry out specific tasks.

This model uses the function of adaptive thinking timerecently presented at the API, to offer three modes of reasoning, to choose between low, medium and high. This way, users can adjust the time the AI ​​takes to think about a problem before offering a response. However, it must be taken into account that, if a low reasoning time, the model responses may suffer from more inaccuracies.

Likewise, as OpenAI has shared, the o3-mini model is capable of offering results comparable to the current o1 reasoning model, however, it obtains these results with a fraction of the computational cost of model o1.

With all this, although the new o3 and o3 mini models have not been officially launched, OpenAI has announced a security researcher program where they can register to obtain a o3-mini preview. Following this line, Altman has announced that they plan to launch the o3-mini model at the end of January 2025 and, after that, the o 3 model.