Over time, mental workload (MWL) has become a major topic of scientific discussions. Nevertheless, there is no reference model involving standardized cognitive tests with distinct levels of difficulty and allowing to consider the correlation between task performance and MWL based on these levels. Therefore, we have developed an innovative protocol to model the variation of self-reported MWL (via the NASA-TLX and Workload Profile questionnaires) for tests requiring different cognitive functions (memory, mental inhibition, mental flexibility, and divided attention). Each of these tasks is an adaptation of existing cognitive tests, for which three equivalent difficulty levels were established in terms of impact on performance. The results showed that our difficulty levels allowed distinct classes of MWL and that it co-varied negatively with the performance across difficulty. Our protocol has thus provided new benchmarks that can be useful in many domains where it’s essential to be able to infer MWL from test performance.