The wide range of spatial and temporal scales present in turbulent flow problems forms a computational bottleneck, which large eddy simulations (LES) circumvent by a coarse graining procedure. The effects of the unresolved fluid motions enter the coarse-grained equations as an unclosed subgrid scale (SGS) term. In recent years, the computational fluid-dynamics community has explored a multitude of data-driven approaches to model the subgrid scale term, typically focusing on reproducing the full correct coarse-grained dynamics. Our approach is grounded in the observation that many practical applications of LES focus on a limited set of quantities of interest (QoIs)—such as average energy or enstrophy—rather than the full flow field. By focusing on these QoIs, we reformulated the SGS modeling task as a low-dimensional learning problem, significantly reducing the computational complexity while improving model interpretability. The key to our approach lies in representing unresolved dynamics using a minimal set of scalar time series, one for each QoI. In recent work, we presented a simple, stochastic, 2D turbulence closure model based on a reduced subgrid scale term. We demonstrated that this method can successfully capture the long-term distributions of four QoIs in a simple 2D turbulence test case, using a random noise model for the time series, provided sufficient training data is available. We now extend this to 3D isotropic turbulence. The simple random noise model no longer suffices as a basis for the SGS term, due to much stronger time correlations in the scalar time series. We will consider more advanced data-driven time series models, learning about key differences between 2D and 3D turbulence in the process.