Thanks to the advancement of knowledge and technologies, we want to simulate larger and larger biological systems. But classical methods must be rethink to be able to cope with such system. At the same time, mainly riding the trends of AI/ML, a novel approach is to use methods mixing different arithmetic precision. Indeed, such method improves arithmetic intensity while saving memory, energy and computational time. However, controlling the errors induced by the mix of different precision remains the major issue. In this paper, we are present a new method allowing the use of mixed precision to solve ordinary differential systems of equations. We evaluate our method against large biological systems. We show that with such systems, we can reduce the arithmetic precision of a part of the biological model will retain the same numerical precision.