Just as the 2025 college entrance exam came to a close, a practical test on whether "AI can solve math college entrance exam questions" was carried out simultaneously. In two rounds of evaluations by professional media, QAQK (Quark) ranked first with scores of 145 and 146 respectively, surpassing DouBao and becoming a new benchmark for domestic AI's mathematical ability.
In the public evaluation, participating models included Quark, DouBao, YuanBao, and ChatGPT, all using the 2025 National College Entrance Exam Math Test Paper (Volume I), with internet search disabled and only deep thinking mode enabled.
In the evaluation by LanJing Finance, Quark successfully topped the list with a score of 145. In terms of multiple-choice and fill-in-the-blank questions, Quark achieved a correct rate of 93%, ranking first. Among the eight single-choice questions, none of the AI products answered question six correctly. Through the chain of reasoning, it was found that the reason for the error was that the AI had a deviation in recognizing vector coordinates and arrow directions in the figure.
In the evaluation by "Four Trees Relative Theory," Quark once again topped the list with a score of 146. In terms of answering speed, Quark ranked first, while DouBao ranked second. For example, Quark could complete one question in 4 minutes, while the average time for other products was around 6 minutes.
The problem-solving capability of Quark is supported by the "Quark Learning Spirit Big Model." This model is based on Tongyi Qianwen, relying on a learning database of billions of materials and post-training capabilities, especially excelling in solving complex scientific problems and bringing users a brand-new experience of heuristic learning.
