This study aims to use molecular dynamics (MD) simulations of Kremer–Grest (KG) chains to inform future developments of models of entangled polymer dynamics. We perform nonequilibrium MD simulations, under shear flow, for well‐entangled KG chains. We study chains of 512 and 1000 KG beads, corresponding to 8 and 15 entanglements, respectively. We compute the linear rheological properties from equilibrium simulations of the stress autocorrelation and obtain from these data the tube model parameters. Under nonlinear shear flow, we compute the shear viscosity, the first and second normal stress differences, and chain contour length. For chains of 512 monomers, we obtain agreement with the results of Cao and Likhtman (ACS Macro. Lett. 2015, 4, 1376). We also compare our nonlinear results with the Graham, Likhtman and Milner‐McLeish (GLaMM) model. We identify some systematic disagreement that becomes larger for the longer chains. We made a comparison of the transient shear stress maximum from our simulations, two nonlinear models and experiments on a wide range of melts and solutions, including polystyrene (PS), polybutadiene, and styrene–butadiene rubber. This comparison establishes that the PS melt data show markedly different behavior to all other melts and solutions and KG simulations reproduce the PS data more closely than either the GLaMM or Xie and Schweizer models. We discuss the performance of these models against the data and simulations. Finally, by imposing a rapid reversing flow, we produce a method to extract the recoverable strain from MD simulations, valid for sufficiently entangled monodisperse polymers. We explore how the resulting data can probe the melt state just before the reversing flow. © 2019 The Authors. Journal of Polymer Science Part B: Polymer Physics published by Wiley Periodicals, Inc. J. Polym. Sci., Part B: Polym. Phys. 2019, 57, 1692–1704