<![CDATA[The ability to solve Higher Order Thinking Skills (HOTS) problems is a critical competency that needs to be developed in 21st-century mathematics education. However, real-world evidence indicates that most students still struggle with solving HOTS problems, particularly in the areas of analysis (C4), evaluation (C5), and creation (C6). This study aims to determine the differences in HOTS problem-solving abilities between students using the Problem-Based Learning (PBL) model and those using the deep learning-based PBL model among 10th-grade students at Al-Islam 1 High School in Surakarta during the 2025/2026 academic year. This study employed a quantitative approach using a quasi-experimental method and a posttest-only nonequivalent groups design. The study population consisted of 267 students, with a sample of 42 students selected using cluster random sampling, comprising Class X.2 as Experimental Group 1 (PBL) and Class X.5 as Experimental Group 2 (deep learning-based PBL). The research instrument consisted of HOTS-type essay questions that had undergone validity and reliability tests with a Cronbach’s Alpha coefficient of 0.880. Data analysis utilized the Shapiro-Wilk normality test, Levene’s Test for homogeneity, and an independent samples t-test for hypothesis testing using IBM SPSS Statistics version 25. The results showed a significant difference in HOTS problem-solving ability between the two classes, with a significance level of 0.040 < α = 0.05. The average score for the deep learning-based PBL class (80.29) was higher than that of the standard PBL class (72.52), indicating that the deep learning-based PBL model is more effective in enhancing students’ ability to solve HOTS-type problems. ABSTRAK Kemampuan penyelesaian soal tipe Higher Order Thinking Skills (HOTS) merupakan kompetensi penting yang perlu dikembangkan dalam pembelajaran matematika abad ke-21. Namun kenyataan di lapangan menunjukkan bahwa sebagian besar siswa masih mengalami kesulitan dalam menyelesaikan soal HOTS, khususnya pada indikator menganalisis (C4), mengevaluasi (C5), dan mencipta (C6). Penelitian ini bertujuan untuk mengetahui perbedaan kemampuan penyelesaian soal tipe HOTS antara siswa yang menggunakan model pembelajaran Problem Based Learning (PBL) dengan siswa yang menggunakan model pembelajaran PBL berbasis deep learning pada siswa kelas X SMA Al-Islam 1 Surakarta tahun pelajaran 2025/2026. Penelitian ini menggunakan pendekatan kuantitatif dengan metode quasi eksperimen dan desain posttest-only nonequivalent groups design. Populasi penelitian berjumlah 267 siswa, dengan sampel sebanyak 42 siswa yang dipilih menggunakan teknik cluster random sampling, terdiri atas kelas X.2 sebagai kelas eksperimen 1 (PBL) dan kelas X.5 sebagai kelas eksperimen 2 (PBL berbasis deep learning). Instrumen penelitian berupa soal uraian bertipe HOTS yang telah melalui uji validitas dan uji reliabilitas dengan koefisien Cronbach's Alpha sebesar 0,880. Analisis data menggunakan uji normalitas Shapiro-Wilk, uji homogenitas Levene's Test, dan uji hipotesis independen sample t-test dengan bantuan IBM SPSS Statistics versi 25. Hasil penelitian menunjukkan bahwa terdapat perbedaan yang signifikan antara kemampuan penyelesaian soal HOTS kedua kelas dengan nilai signifikansi 0,040 < α = 0,05. Rata-rata nilai kelas PBL berbasis deep learning (80,29) lebih tinggi dibandingkan kelas PBL (72,52), sehingga model PBL berbasis deep learning terbukti lebih unggul dalam meningkatkan kemampuan penyelesaian soal tipe HOTS siswa. ]]>
