The next version of OpenAI for o3 will arrive early next year

After nearly two weeks of announcements, OpenAI released its 12 Days of OpenAI series with a preview of the next generation. “In honor of our friends at Telefónica (owner of the O2 cellular network in Europe), and in the grand tradition of OpenAI being the worst at names, it’s called o3,” OpenAI CEO Sam Altman told viewers. announcement on YouTube.

This new version is not ready for public use. Instead, OpenAI is starting to make o3 available to researchers who want help security testing. OpenAI also announced the existence of o3-mini. Altman said the company plans to launch the brand “in late January,” with o3 following “soon.”

As you might expect, the o3 offers better performance than its predecessor, but how it’s better than the o1 is the headline here. For example, when they pass this year American Invitational Mathematics Examinationo3 achieved an accuracy rate of 96.7 percent. In contrast, o1 scored the lowest percentage of 83.3. “What that means is that o3 often misses one question,” said Mark Chen, vice president of research at OpenAI. In fact, o3 performed so well in the benchmarks OpenAI puts its models that the company had to find a difficult test to prove it.

One of them is ARC-AGIa benchmark that measures AI’s ability to learn and learn on the fly. According to the developer of the test, to no avail ARC awardan AI system capable of defeating ARC-AGI could represent “the most important thing in modern intelligence.” Since its launch in 2019, no AI model has beaten ARC-AGI. The test consists of questions that most people can get by experience. For example, in the example above, the correct solution would be to make squares out of four polyominoes using black blocks.

In its lowest ranking, the o3 scored 75.7 percent on the test. With the added power, the model achieved a rating of 87.5 percent. “People’s participation equals 85 percent, so to be on top of that is a big milestone,” said Greg Kamradt, president of the ARC Prize Foundation.

OpenAI also introduced the o3-mini. The new model uses OpenAI’s recently announced Adaptive Thinking Time API to provide three types of thinking: Low, Medium and High. Instead, this allows users to change how long the program “thinks” about a problem before providing a solution. As you can see in the graph above, o3-mini can achieve similar results to the current OpenAI proposal o1, but at a computational cost. As mentioned, the o3-mini will arrive for public use ahead of the o3.