Allele mining of crop pangenomes can enable the identification of novel variants for trait improvement, increase crop genetic diversity, and purge deleterious mutations around fixed genomic regions. Sorghum, a C4 cereal crop domesticated in the tropics, was selected for reduced plant height and maturity to develop combine-harvestable and photoperiod-insensitive US grain sorghums. To breed semi-dwarf US grain sorghum hybrids, public and private sector programs mostly used the dw3-ref allele, which produces undesirable height revertants (frequency of 0.1-0.3%) due to uneven crossing over at the 882 bp tandem duplication region. Although the dw3-ref allele produces revertants, US sorghum breeding programs continued using this allele in the absence of identified allelic variants that suppress revertants. In this study, we leveraged a sorghum pangenome resource (resequenced sorghum association panel and a global diversity panel of 1661 lines) to identify seven loss-of-function variants in the Dw3 gene using the SnpEff variant calling prediction. We validated the Segaolane dw3 loss-of-function variant, resulting from a 137 bp deletion in the third exon, to suppress revertant production. Segaolane NAM family RILs with the dw3-ref allele produced revertants while no revertants were observed in RILs with the Segaolane dw3 allele. The availability of resequencing data enabled the designing of haplotype-based markers detecting the Segaolane stable dw3 allele for marker-assisted trait introgression into elite sorghum breeding lines. This research mining new stable-dwarfing dw3 alleles demonstrated the application of sorghum pan-genome for trait improvement and developing marker-assisted breeding strategies.