According to data from ParallelKernelBench, it has been revealed that GPT-5.5 and similar models are facing challenges when dealing with multi-GPU CUDA kernels. They are reportedly only able to successfully solve less than 31% of the tasks assigned to them. This performance issue has significant implications in terms of efficiency and effectiveness in handling tasks that require parallel processing power. Further insights on this matter can be found in the complete article.