Specializing models for K-12 education does not require a broad loss of general capability, with MMMU-Pro accuracy increasing at all three scales¶

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? MMMU-Pro overall accuracy increases from 50.46% to 52.60% for 4B, from 58.38% to 60.75% for 9B, and from 64.97% to 67.98% for 27B after education-oriented tuning. → Hao Liang 2026 q2 i? Across the three scales, tuning generally improves IFEval metrics and improves both GPQA Main and Diamond accuracy. → Hao Liang 2026 (2)

Evidence¶

Hao Liang 2026¶

Hao Liang, Qihan Lin, Meiyi Qiang, Linzhuang Sun, Hengyi Feng, Mingrui Chen, Sizhe Qiu, Wentao Zhang. (2026). OmniEdu: Open Foundation Models for Learning and Teaching. https://arxiv.org/abs/2609.23088 doi:10.48550/arxiv.2609.23088

q2 · i? · design · r2

General-capability retention evaluation (Section 4.5) of base versus tuned models on IFEval, GPQA, and MMMU-Pro. The article reports MMMU-Pro overall accuracy increases at all three scales, though changes vary by MMMU-Pro discipline.

"MMMU-Pro overall accuracy increases from 50.46% to 52.60% for 4B, from 58.38% to 60.75% for 9B, and from 64.97% to 67.98% for 27B."

Hao Liang 2026 (2)¶

Hao Liang, Qihan Lin, Meiyi Qiang, Linzhuang Sun, Hengyi Feng, Mingrui Chen, Sizhe Qiu, Wentao Zhang. (2026). OmniEdu: Open Foundation Models for Learning and Teaching. https://arxiv.org/abs/2609.23088

q2 · i? · design · r2

Section 4.5 auxiliary diagnostics of instruction following (IFEval) and scientific reasoning (GPQA). The article reports tuning "generally improves IFEval metrics and improves both GPQA Main and Diamond accuracy".

"Across the three scales, tuning generally improves IFEval metrics and improves both GPQA Main and Diamond accuracy."

Discussion¶

Related Claims¶

• Education-oriented instruction tuning consistently improves curriculum grounding, K-12 problem solving, and pedagogical tutoring at every evaluated model scale — a broader claim this one bears on
• Education-oriented tuning improves authentic K-12 problem solving, with OmniEdu-27B's MDK12-Bench open-question correctness rising from 62.08% to 78.77% — related
• LoRA adaptation significantly increases closed-book accuracy on the PhysicsMate test set at all three Qwen3 scales — related
• A single difficulty-weighted SFT run on a vision-language base model reaches 61.64% overall accuracy on TutorBench, surpassing Gemini-3 Pro (58.52%) — 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