The legacy subgroup method of the APOLLO3® code relies on the use of mathematical probability tables and on the General Resonance model. It yields precise multiplication factors and reaction rates on a fine 383 energy group mesh for typical light water reactor calculations, at the cost of an expensive computational time. Recently, a subgroup method based on the physical probability tables and on the Intermediate Resonance model has been implemented in APOLLO3® in an effort to reduce the number of energy groups required to achieve a similar accuracy down to 69. This paper analyses the newly developed method by comparing it to the legacy one on simplified numerical test cases of 3 × 3 UO2 pin-cells, with a central cell composed of either a water hole or a Gd-UO2 pin-cell. The difference of multiplication factor on these cases is less than 250 pcm for both subgroup methods. The newly implemented subgroup method displays errors twice as big as the legacy subgroup method, but with an acceleration of the self-shielding and of the flux calculations of a factor 12 and 5, respectively.