Background: Artificial intelligence (AI)–based curriculum personalization may reduce access barriers and respond to learner variability, yet quasi-experimental evidence in inclusive school education remains scarce. Objective: To evaluate the effect of a real-time adaptive model on curriculum personalization among students with cognitive and sensory diversity in tenth grade of basic education and first year of upper-secondary education. Method: A quasi-experimental pretest–posttest study included 200 students equally allocated to intervention and comparison conditions. The intervention group completed 30 mathematics sessions in which the system adjusted difficulty, pace, format, and feedback; the comparison group used non-adaptive digital resources. Change scores, effect sizes, and linear models adjusted for pretest, educational level, diversity profile, age, attendance, digital competence, and support level were estimated. Results: Personalization gains were greater in the intervention group (M = 0.692, SD = 0.313) than in the comparison group (M = 0.202, SD = 0.239), difference = 0.489, 95% CI [0.411, 0.567], d = 1.76, p < .001. The adjusted effect was b = 0.481, 95% CI [0.401, 0.561]. Advantages were also found for achievement (d = 2.03), engagement (d = 1.29), accessibility (d = 1.86), and participation (d = 1.94), all p < .001. No significant moderation by educational level or diversity profile emerged. Conclusion: The adaptive model was associated with broad and consistent improvements; nevertheless, non-random allocation and unavailable cluster identifiers limit causal inference.