Estimating the long-term stability of biochar in soil often relies on extrapolating mineralization data from short-term laboratory incubations. Various models such as single first-order (SFO), double first-order (DFO), triple first-order (TFO) and the power model have been employed for this purpose, all of which have an inherent assumption that biochar is completely biodegradable. However, recent insights challenge this assumption by highlighting that biochar consist largely of highly condensed aromatic structures, which have been proposed to be essentially inert. If biochar were resistant to microbial degradation it would make sense if this was reflected in the choice of model used. Therefore, our aim was to assess whether the proposed inert pool models (SFO+I and DFO+I) fit the data better compared to existing models (SFO, DFO, TFO and power model) using a recently compiled extensive dataset. We hypothesized that models incorporating an inert pool would fit better (or at least comparably well) to incubation data compared to the existing models and give more reliable long-term predictions. As a way of assessing the model’s predictive ability, we fitted them to progressively shortened incubation times derived from the longest biochar incubation data sets available, and then evaluated how well they extrapolated to the full measured range. Our results indicated that the proposed DFO+I model did indeed fit better than the both DFO and TFO models. Moreover, predictions of BC100 (% of carbon remaining after 100 years) by the inert pool models and by the power model displayed stronger correlations with the biochar stability indicator (H/C ratio) than both SFO and DFO models, which aligns with our initial hypothesis. However, the power model in general outperformed all other models, including the inert pool models, with the highest number of best fits. From our extrapolation exercise, it is clear the DFO model, which has been most widely used to date, substantially underestimated biochar stability in the longer term while the inert pool models tended to overestimate it. This uncertainty appears to be quite severe when fitting inert pool models to incubation data from non-pyrolysed materials. By comparison, the power model appeared to be more robust when estimating biochar persistence in soils. From these results we cannot conclusively confirm nor reject the idea of inert pool models. It is possible that an inert-pool model is the more suitable choice for extrapolating biochar decomposition data. However, it is clear from their tendency to overestimate stability of biodegradable materials. Our current understanding is hampered by the fact that the incubation data set available contains a high proportion of biochars produced at relatively modest temperatures and by a lack of chemical/structural characterization of the incubated biochars. Future research could remedy this by providing better information on the degree of aromatic condensation/proportion of inertinite in incubated biochars samples. For now, we recommend the power model as the most robust option.