VietEduQwen improves over its base model across all eight exam subjects, with the largest gains in STEM disciplines, and its residual errors concentrate in multi-step computation and cross-concept integration¶

Claim · All claims
Evidence · 1 study (2 entries) · design r2 · q2
Currency · newest study 2026 · not yet checked against its sources · not reviewed by a person · recheck by 2029: tool and model accuracy changes quickly

Subclaims¶

q2 i? VietEduQwen consistently improves over Qwen3-8B across all eight subjects, with the largest gains in Mathematics, Physics, and Chemistry. → Thang Doan Viet 2026 q2 i? Of 200 sampled errors, approximately 41% involve multi-step numerical computation and 33% involve integration across two or more textbook concepts. → Thang Doan Viet 2026 (2)

Evidence¶

Thang Doan Viet 2026¶

Thang Doan Viet, Anh Nguyen Hoang, Tinh Luong Son, Anh Hoang Thi Ngoc, Huyen Giang Thi Thu, Tai Le Quy. (2026). ConnectED: A Curriculum-Aligned AI System for Vietnamese Instructional Lesson Planning and Student Learning. arXiv. https://arxiv.org/abs/2607.28647 doi:10.48550/arxiv.2607.28647

q2 · i? · design · r2

Subject-wise accuracy analysis across eight examination subjects, presented in Fig. 7. The text reports the model "consistently improves over the base Qwen3-8B model across all subjects" and narrows the gap with commercial baselines in STEM.

"VietEduQwen consistently improves over the base Qwen3-8B model across all subjects. The largest gains are observed in STEM disciplines, particularly Mathematics, Physics, and Chemistry, where VietEduQwen substantially narrows the performance gap with Gemini-2.5 and Gemini-3 variants."

Thang Doan Viet 2026 (2)¶

Thang Doan Viet, Anh Nguyen Hoang, Tinh Luong Son, Anh Hoang Thi Ngoc, Huyen Giang Thi Thu, Tai Le Quy. (2026). ConnectED: A Curriculum-Aligned AI System for Vietnamese Instructional Lesson Planning and Student Learning. arXiv. https://arxiv.org/abs/2607.28647

q2 · i? · design · r2

Manual error analysis of a random sample of 200 errors from VietEduQwen's incorrect predictions. The authors report "Approximately 41% involve multi-step numerical computation" and a further 33% cross-concept integration failures; the remaining 26% span reading comprehension, unit conversion, and ambiguous distractors.

"Approximately 41% involve multi-step numerical computation where intermediate rounding or sign errors propagate to incorrect final answers. A further 33% involve questions requiring integration of information across two or more textbook concepts"

Discussion¶

Related Claims¶

• VietEduQwen achieves 87.02% accuracy on the 2025 Vietnamese National High School Examination, a 6.10-percentage-point gain over the base Qwen3-8B model — a broader claim this one bears on
• LLM teaching effectiveness varies substantially across subject domains, strongest where knowledge points apply directly and weakest in physics and chemistry — related
• Removing the DPO objective reduces examination accuracy by 1.11 percentage points and degrades annotated pedagogical quality — related