TACTutor shows no degradation relative to its Qwen backbone on external educational benchmarks, with scores rising on MRBench, TutorBench, LongTutor, and DeepTutor¶

Claim · All claims
Evidence · 1 study · 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? On external transfer checks, TACTutor shows no degradation relative to the Qwen3.5-4B backbone: MRBench macroDAMR rises from 0.743 to 0.880, TutorBench weighted score from 0.497 to 0.519, LongTutor teaching quality from 3.712 to 3.834, and the DeepTutor smoke score from 3.565 to 3.627. → Dongjie Yang 2026

Evidence¶

Dongjie Yang 2026¶

Dongjie Yang, Siyan Lin, Leixian Shen, Rui Sheng, Huamin Qu, and Zixin Chen. (2026). TACT: Taxonomy-Aligned Post-Training for Pedagogically Adaptive English Tutoring. https://arxiv.org/abs/2608.03952 doi:10.48550/arxiv.2608.03952

q2 · i? · design · r2

Transfer checks on external educational benchmarks covering math tutoring, STEM tutoring, long-history personalization, and interactive tutor simulation (Appendix Table 3). The article reports "no degradation relative to the corresponding Qwen baselines" with the printed score changes; no significance tests are reported.

"TACTutor shows no degradation relative to the corre-sponding Qwen baselines: MRBench (Maurya et al. 2025) macroDAMRrisesfrom0.743to0.880,TutorBench(Srini-vasa et al. 2025) weighted score from 0.497 to 0.519, Long- Tutor (Li et al. 2026) teaching quality from 3.712 to 3.834, and the DeepTutor (Zhao et al. 2026) smoke score from 3.565 to 3.627."

Discussion¶

Related Claims¶

• In a blinded study with 50 learners, TACTutor receives the highest overall mean rating (5.54/7) and is rated above its backbone across all four teaching-behavior dimensions (+0.39 overall) — related
• Education-oriented tuning produces the largest gains in scaffolded instruction and use of longitudinal student evidence, with LongTutor Evidence averages rising sharply at all scales — related
• Taxonomy-guided post-training raises TACTutor's TACT_Overall on TACTBench from 0.629 (backbone) to 0.832, the highest among evaluated models — related
• Post-training improves TACTutor's auxiliary guardrail flags on TACTBench: accept rate rises from 0.603 to 0.872 and leak-or-overhelp falls from 0.346 to 0.026 — related
• DeepTutor improves overall first-person interactive tutoring quality by 10.76% over a Naive Tutor baseline on TutorBench — related
• Education-oriented instruction tuning consistently improves curriculum grounding, K-12 problem solving, and pedagogical tutoring at every evaluated model scale — related
• The investigate–solve–write scaffold, with personalization disabled, improves all five backbone model families on general agentic benchmarks, with average relative gains of 25.69%–32.03% — related
• In the public TutorBench sample, agency-oriented rubric criteria are most concentrated in the active-learning use case (agency-dimension density 0.720, agency-skill density 0.411) — related