AI beats humans in the most difficult math test it has ever been subjected to. Four ender models chatgpt 5.5 Pro were put to the test and none of them managed to answer 10 questions correctly.

According to the article published in the scientific journal Nature, the highest result was achieved by the Zurich Polytechnic Eth model, which solved 6 problems out of 10.

ChatGpt and OpenAi came in third and fourth place, while the Princeton model came last, unlike others are based on Google’s Gemini.

The test performed is First Proof, an independent project promoted to follow the evolution of AI capabilities in mathematical research.

The questions were never published before either in the scientific literature or on the Internet, to reduce the risk that the models copied, or were part of the early learning phase.

The test shows that all the models lack intuition and the ability to connect the elements leading to the final solution, so the basic way was correct, but the IA systems failed to capture all the details, which makes them limited compared to the human and his intuition, in other words, the sixth sense.

Source: prizrenpost