Computational fluid dynamics represent a powerful tool to assess the performance of a combustor and identify possible issues/instabilities, helping thus e.g. to investigate the impact of advanced cycle modifications on the combustion in mGTs. The steady Reynolds-averaged Navier-Stokes (RANS) approach is still mostly used in this framework. With growing computational power, Large Eddy Simulation (LES) has gained more interest. LES provides higher details concerning flow structures and can better predict possible instabilities, specifically needed for advanced cycle modelling. On the other hand, LES remain rather challenging for real industrial applications. This work aims at providing an answer whether the advantages of LES justify the much higher computational costs. The objective of the present study is thus to assess the combustion performance and emissions of a typical small-scale 3.2 kWe micro gas turbine (mGT), using steady RANS and LES for various fuels. In this framework, a comparison of RANS and LES approaches (two levels of fidelity) is performed on a typical industrial case, to point out the strengths and weakness of each method with regard to industrial and research needs. The results show that both RANS (at a reduced cost) and LES can accurately predict the time-averaged trends of the main performance parameters, like temperature levels and emissions, also using various non-conventional inlet conditions. For the accurate prediction of the instabilities, the LES approach stands out as this approach takes into account the time-variation of the different quantities. Finally, a significant discrepancy has been observed between the CO levels provided by RANS and LES approaches where LES is overestimating the level of CO in the exhaust gases. Whereas it is difficult for LES to compete with convincing results provided by RANS, especially in the prediction of global emissions at reduced simulation cost, the LES strengths come out especially in flame and combustion stability analysis.