It found that standard AI models outperformed LRMs in low-complexity tasks, while both types of model suffered “complete collapse” with high-complexity tasks. Large reasoning models attempt to solve complex queries by generating detailed thinking processes that break down the problem into smaller steps.