The E3SMv2.1 seasonal-to-multiyear large ensemble (E3SMv2.1-SMYLE) forecast system is a collection of 20-member, 28-month hindcasts initialized every November 1^st and May 1^st from 1970 through 2019. The experimental design and initialization methodology follow the template of a similar system that uses the CESM2 model (CESM2-SMYLE). We present a global assessment of prediction performance for seasonal surface climate that, to first order, reveals a rough equivalence of the E3SMv2.1- and CESM2-based systems. Both systems exhibit competitive skill at predicting DJF Niño3.4 when initialized in the preceding year, and E3SM shows slightly improved skill at multiyear lead times. The skill comparison for regional surface air temperature, precipitation, and sea level pressure is very spatially and temporally heterogeneous, with neither system showing consistent improvement over the other. The combined 40-member prediction system, on the other hand, is found to be generally superior to either of the single-model prediction systems. The E3SMv2.1-SMYLE dataset will facilitate a broad spectrum of research on the predictability of climate extremes from seasonal to multiyear timescales while also permitting informative comparisons between E3SMv2.1 and CESM2 that could benefit both DOE and NSF efforts to understand Earth system predictability across these timescales.