A novel modeling technology termed "Data Physics" that combines reservoir physics with machine learning is validated against a conventional reservoir simulator for thermal recovery problems, i.e., steam flooding and steam assisted gravity drainage (SAGD). The novelty of the new model is its combination of speed of data integration (less than a week) and runtime (minutes) with long-term predictive accuracy (years or decades). This is due to the unique integration of reservoir physics with fast data-driven methods. For accurate benchmarking, major sources of modeling errors in the finite difference simulations are screened and controlled. Two cases are studied, the SPE4 steamflood model, and a single pad SAGD model. The results demonstrate that the Data Physics model is able to reproduce production profiles and key reservoir physics accurately when numerical errors in simulation are properly accounted for, while also being immune to numerical issues like grid orientation effects that can have significant impact on results of reservoir simulation.