To simultaneously consider mean yield and interannual stability in cotton variety evaluation, this study integrated yield level and stability into a unified framework. Ten cotton varieties were evaluated using four years (2022–2025) of field trial data. Mean seed cotton yield, coefficient of variation (CV), and Shukla’s stability variance characterized yield level, interannual fluctuation, and variety × year interaction stability, respectively. The Finlay–Wilkinson regression coefficient assisted in analyzing varietal responses to year-specific environments, and entropy weight–TOPSIS was used for comprehensive ranking. Year, variety, and variety × year interaction significantly affected seed cotton yield (p < 0.001). Weights of CV, mean yield, and Shukla’s stability variance were 42.3%, 32.1%, and 25.7%, respectively. Mean-yield ranking was generally consistent with the TOPSIS ranking, though some varieties changed. CCRI 100 had a mean yield of 3413 kg ha⁻¹ and a CV of 5.4%. It ranked first in both the mean-yield ranking and the entropy weight–TOPSIS comprehensive ranking. SCRC 28 ranked first in stability but lower in yield, ranking second overall. This study demonstrates that entropy weight–TOPSIS can integrate yield level, interannual fluctuation, and variety × year interaction stability into a single evaluation framework, providing an effective method for shifting from “high-yield screening” to “coordinated high-yield and stable-yield screening” in multi-year variety trials.