This study aimed to develop and evaluate a longitudinal risk prediction model for recidivism among offenders with intellectual disabilities, integrating static and time-varying risk and protective factors, and to compare its predictive performance with a conventional cross-sectional model. A longitudinal cohort design was used, comprising 1,842 offenders with documented intellectual disability identified through linked criminal justice and disability registers. Participants were followed for a median of 4.8 years (IQR: 2.6–6.9), contributing 8,412 person-years of observation. Recidivism was defined as a new criminal justice concept. Cox proportional hazards models with time-dependent covariates were applied. The concordance index (C-index), time-dependent area under the curve (AUC), and calibration slope were used to evaluate model performance, and internal validation was done through bootstrapping. Follow-up showed 712 individuals (38.7%) reoffended, resulting in an incidence rate of 8.5 events per 100 person-years. The younger age (HR = 0.97 per year, 95% CI: 0.96-0.98) and the previous criminal history (HR = 1.18/offense, 95% CI: 1.14-1.22) were considered as the static predictors. Mild ID was linked to the increased risk of recidivism (HR = 1.42, 95% CI: 1.19-1.69). Mental instability over time (HR = 1.61) was also a risk factor, and treatment attendance (HR = 0.66) and good housing (HR = 0.71) were protective. The longitudinal model was more effective with the C-index = 0.73; AUCs = 0.76 at 1 year, 0.74 at 3, and 0.72 at 5 years, respectively, than the cross-sectional model (C-index = 0.64). Longitudinal models that include dynamic risk and protective variables outperform traditional recidivism prediction with offenders with intellectual disabilities and help focus on personalized and evidence-based risk management approaches.