Predicting weather accurately at Earth's surface requires properly representing all that lies above the surface. This includes the stratosphere, where variations in winds and temperature help forecast winter weather weeks to months in advance. I am collaborating with the European Centre for Medium-Range Weather Forecasts (ECMWF) to evaluate and improve the representation of the stratosphere in their widely used atmospheric models. ECMWF develops reanalyses, which combine Earth observations and atmospheric modelling to produce global, historical, gridded databases of atmospheric variables such as temperature and humidity. ERA5 is their current reanalysis version; it was released in 2018 and has over 100,000 users. ECMWF will release their next reanalysis version, ERA6, in 2027. I am characterizing how the representation of the stratosphere has evolved between ERA5 and ERA6 and understanding which developments in the modelling system have contributed to the changes that I see. My preliminary analysis shows substantial improvement in stratospheric horizontal winds in ERA6, especially in the tropical stratosphere, which is particularly poorly represented in ERA5. Since ERA6 is not yet available, I am comparing a yearlong ERA6 prototype experiment (October 2019-September 2020) with the same period from ERA5. I am evaluating both reanalyses against coincident Earth observations. Some of those Earth observations have been used to produce the reanalyses and others have never been seen by the modelling systems and thus support independent tests of model improvement. I am particularly interested in isolating the impacts of a discrete change made to the model physics between ERA5 and ERA6: the reduction of turbulent diffusion in the lower stratosphere. When ECMWF scientists change the model physics, they have to run control and experimental simulations in both the summer and the winter to test that change. I will also evaluate these "turbulence test" simulations, which span June-August 2020 (overlapping the ERA6 prototype experiment) and December 2020-February 2021. Finally, I want to understand what caused the stratosphere to change in ERA6. The major possibilities are 1) changes to the assimilation of Earth observations, 2) changes to the model physics, and 3) increases in horizontal resolution in ERA6. Reanalyses contain analyses, which integrate information from Earth observations each hour, and forecasts, which run for a maximum of 18 hours without integrating information from Earth observations after the initial timestep. I will compare the atmospheric state at the end of the forecast simulations between ERA5 and ERA6. There, the influence of Earth observations is reduced and the influence of differences in model physics is more visible. In order to investigate 3, I will evaluate the ERA6 prototype at the resolution of ERA5 and at its own higher resolution. This work will quantify improvement in the stratosphere in ERA6 and highlight areas that are still in need of improvement. This is of immediate relevance to the 100,000 users of ECMWF's reanalyses. I will also identify the causes of improvement, which will help ECMWF and other modelling centers steer their future model development towards achieving a more realistic stratosphere.