Abstract: This research addresses the unreliability of large language models in the field of mechanics by proposing a “knowledge-computation-process” triple framework. Based on this framework, the general theoretical mechanics problem-solving system (GT-Mech) was developed. To test the framework's effectiveness, GT-Mech participated as the first AI contestant in the 15th National Zhou Peiyuan Competition on Mechanics for College Students, achieving a performance equivalent to the Special Prize level. This study used multiple sets of final exam papers for theoretical mechanics from Tsinghua University to conduct comparative tests between GT-Mech and domestic general baseline models, with GT-Mech's scores approaching an excellent level. This research not only validates the potential of AI in tackling complex mechanics problems but also provides a feasible technical paradigm for building trustworthy, domain-specific artificial intelligence systems. It holds significant reference value for promoting the deep integration of AI in mechanics and other basic sciences, as well as for advancing intelligent education.