Inclusive education in Indonesia continues to face persistent gaps in meeting the needs of students with special educational needs (SEN), primarily due to the absence of a systematic teaching strategy that integrates deep learning and Culturally Responsive Teaching (CRT). This study aims to develop and validate the SOCIAL-L DLCRT model for inclusive primary schools using a Research and Development (R&D) design with the ADDIE model, involving 104 teachers and 37 SEN students from eight schools in Bogor City. Data were collected through observations, FGDs, interviews, questionnaires, and HOTS tests, and analyzed using Aiken's V, paired t-tests, N-Gain, and Cohen's d, with retention measured one month post-intervention. Needs analysis revealed that 78% of teachers had never received CRT training, deep learning implementation was low (mean = 2.08), and 41% of SEN students were passive. The SOCIAL-L DLCRT model (stages: S1, O, S2, I, A, L) achieved high validity (mean Aiken's V = 0.86). Implementation with 37 SEN students significantly increased HOTS scores from 56.2 to 78.4 (l = 16.31; p < 0.001; Cohen's d = 2.68; N-Gain = 0.499, medium category), increased active participation by 32 to 42 percentage points, and resulted in 97.2% learning retention after one month, with positive responses from teachers (91.8%) and SEN students (88%), as well as an implementation fidelity of 87.5% (very high category). In conclusion, the SOCIAL-L DLCRT model is valid, effective, and sustainable as a CRT-based deep learning optimization strategy contributing to SDG 4, although limitations include its geographical focus on Bogor and the absence of a control group; therefore, further research is recommended in different cultural contexts, as well as the development of digital modules based on Universal Design for Learning (UDL).