Key message <p><b>Designing genomic selection to capture the additive and epistatic effects.</b></p> Abstract <p>Genomic selection (GS) offers great potential to accelerate long-term genetic gain, but strategic decisions such as progeny size and number of crosses remain poorly established, particularly under contrasting resource scenarios. We conducted stochastic simulations of rice breeding programs over 50&#xa0;years (10 cycles) using progeny sizes of 25, 50, 100, and 200 individuals, under both theoretical (unlimited resources) and practical (budget-constrained to 4000 F₂ individuals) contexts, and considering three levels of epistasis (absent, moderate, high). In theoretical scenarios, larger progenies consistently achieved higher gains. After 50&#xa0;years, progenies of 200 individuals reached cumulative responses to selection of 2.39 (1.96%&#xa0;yr⁻¹) with no epistasis, 3.20 (2.60%&#xa0;yr⁻¹) under moderate epistasis, and 3.48 (3.34%&#xa0;yr⁻¹) under high epistasis. These schemes also maximized prediction accuracy and efficiently converted additive and epistatic variance into genetic gain. Conversely, under budget constraints, smaller progenies combined with more crosses outperformed larger ones. Progenies of 25 and 50 individuals achieved the greatest responses—up to 2.58 (2.07%&#xa0;yr⁻¹) without epistasis, 3.36 (2.76%&#xa0;yr⁻¹) under moderate epistasis, and 2.72 (2.45%&#xa0;yr⁻¹) under high epistasis—while maintaining higher genetic diversity across cycles. Our results demonstrate that in resource-unlimited conditions, larger progenies (200 individuals) maximize the capture of additive and epistatic effects, whereas in budget-constrained programs, smaller progenies (25–50 individuals) coupled with more crosses provide the most efficient strategy. These findings provide practical guidelines for breeders to design GS schemes that reconcile high long-term genetic gain with operational feasibility, highlighting the decisive role of epistasis in shaping gain trajectories.</p>