Modelling and predicting soil reinforcement by roots is a central quest in eco-engineering to assess the effectiveness of vegetation in landslide or erosion mitigation. Several fibre bundle model algorithms (FBMs) have been successfully introduced for soil reinforcement estimation and have achieved satisfactory results. However, FBMs yield variable reinforcement estimates due to their different hypotheses on the progressive root failure mode in soil (e.g., failure driven by load, strength or strain of roots); thus, they lack consensus in mechanism and application. Herein, we proposed three new FBMs, namely, FBMs-W, in which the root failure mode is driven by energy (i.e., work), and compared them with others using ground truth data. These FBMs-W allow, to some extent, reconciliation of the previous assumptions in conflicts because both load and displacement of roots affect in root failure procedures. To measure soil reinforcement, we conducted in situ direct shear tests under both rooted and bare soil conditions and two soil moisture levels using living roots from four common tree or shrubby species (Robinia pseudoacacia L., Pinus tabulaeformis Carr., Syringa oblate Lindl., and Vitex negundo L. var. heterophylla (Franch.) Rehd.) on the Loess Plateau, China. The observed soil reinforcement was then compared with predicted ones using different FBMs for model assessment. All the FBMs overestimated the soil reinforcement compared with the observed values. Among the FBMs, the three FBMs-W showed fairly conservative predictions of root cohesion (cr) that were close to each other. FBMs-W gave the most stable soil reinforcement estimates with the smallest standard deviations. The above results were consistent among the four species. Our findings highlight the high disparities in soil reinforcement estimates and bias from the observed values due to the choice of model algorithms. Due to the conservativeness and stability of the soil reinforcement estimates and good coherence with the observed root failure order, we recommend that FBMs-W should be good candidates to choose by future modellers and practitioners when evaluating the likelihood of the stability of a vegetated slope.