Objective. Reliable myocardial T1 mapping is essential for quantitative cardiac tissue characterization but remains sensitive to respiratory motion, particularly during free-breathing acquisitions. This study evaluated the test–retest repeatability of MBSS-T1, a subject-specific, model-based, self-supervised deep-learning method for motion-robust myocardial T1 mapping. Approach. Seventeen healthy male volunteers aged 20–36 years underwent 3 T cardiac MRI using a 5(3)3 MOLLI acquisition under breath-hold (BH) and free-breathing (FB) conditions. Same-day repeat scans were acquired after a 20 min interval that included removal from and repositioning in the scanner. T1 maps were reconstructed using (1) MBSS-T1, (2) the contrast-agnostic deep-learning registration method SynthMorph, (3) scanner-inline motion correction, and (4) no motion correction. Repeatability was evaluated using intraclass correlation coefficients (ICC), relative Bland–Altman analysis, within-subject coefficients of variation (CVw), relative 95% repeatability coefficients (RC95), and variance-based bootstrap inference with Holm correction. Main results. All four methods included 17 complete participant-level test–retest pairs under each breathing condition. Under BH conditions, the global ICC point estimates were 0.77, 0.64, 0.65, and 0.60 for MBSS-T1, SynthMorph, scanner-inline correction, and no correction, respectively. Under FB conditions, the corresponding values were 0.41, 0.32, 0.27, and 0.16. MBSS-T1 also had the lowest descriptive CVw under both conditions. However, none of the paired comparisons of absolute test–retest differences was significant after Holm correction (all adjusted p ≥ 0.623). The smallest Holm-adjusted variance-ratio result was obtained for MBSS-T1 versus no correction under FB conditions (variance ratio 0.43, 95% CI 0.30–0.68; adjusted p = 0.058), which did not meet the prespecified significance threshold. Significance. MBSS-T1 yielded favourable descriptive repeatability estimates, particularly under FB conditions, but no pairwise comparison remained statistically significant after correction for multiple comparisons. These results characterize same-day, same-scanner measurement precision rather than absolute T1 accuracy and require confirmation in larger and more clinically diverse cohorts.