CDPK performance scales with model size, with a sharp Pareto-frontier drop-off below around 8B parameters¶

Claim · All claims
Evidence · 1 study · associational r2 · q2
Currency · newest study 2025 · not yet checked against its sources · not reviewed by a person · recheck by 2028: tool and model accuracy changes quickly

Subclaims¶

q2 i? Among open-weight models with known parameter counts, CDPK accuracy scales with size; the Pareto efficiency frontier shows a sharp drop-off below around 8B parameters, though at 7B parameters performance still spans over 20 percentage points (46% to 66%). → Lelièvre 2025

Evidence¶

Lelièvre 2025¶

Lelièvre, M., Waldock, A., Liu, M., Valdes Aspillaga, N., Mackintosh, A., Ogando Portela, M. J., Lee, J., Atherton, P., Ince, R. A. A., & Garrod, O. G. B. (2025). Benchmarking the Pedagogical Knowledge of Large Language Models. https://arxiv.org/abs/2506.18710 doi:10.48550/arxiv.2506.18710

q2 · i? · associational · r2

Numerical analysis of CDPK accuracy against parameter count for open-weight models only, shown in Figure 9. The article reports the sharp drop-off below around 8B parameters and the 46%–66% range at 7B parameters; the largest model is not the best performing although the smallest is the worst.

"The Pareto efficiency frontier shows a sharp drop-off below around 8B parameters. However, even at 7B parameters there is a range of performance of over 20 percentage points (from 46% to 66%)."

Discussion¶

Related Claims¶

• LLM accuracy on the CDPK pedagogy benchmark ranges from 28% to 89%, with closed-weight reasoning models dominating the top of the leaderboard — related
• The benchmark procedure is stable under stochastic decoding, with a maximum accuracy standard deviation of 0.43% over 20 repeated runs — related
• Accuracy-cost value frontier: at $0.10 per million tokens, CDPK performance rose from 50% (April 2024) to 82% (June 2025) — related
• Under identical LoRA hyperparameters and training data, the 27B Gemma model outperforms the 70B LLaMA model on every computed essay-scoring metric — related
• Larger Llama Guard models outperform smaller ones on education-prompt classification, with the 8B model best in accuracy, recall, and F1 — related
• 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 — related
• Model family and instruction-tuning approach appear better predictors of tutoring quality than parameter count alone — related
• Model size is not the primary factor in outcome-prediction performance — reports the opposite
• Model scale does not consistently improve pedagogical safety: within the Qwen2.5 family the 72B model beats the 7B on only 17 of 33 subject-dimension pairs — related
• Model size did not correspond to higher scores in secondary science; small open-weight models outperformed significantly larger ones — reports the opposite
• Adaptation gains grow monotonically with model capacity under the same LoRA recipe, though three data points cannot establish a scaling law — a narrower finding that bears on this claim
• The domain-specialized 32-billion-parameter model surpasses significantly larger general-purpose models such as Qwen3-235B-A22B-Thinking (86.63%) and Qwen2.5-72B-Instruct (78.60%) on the CDPK Benchmark — related
• In the fine-tuning experiments, increasing base model size up to 32B parameters improved performance, and larger LLMs consistently outperformed smaller MLMs — related
• Over a quarter of benchmark tasks (23 of 80) defeat all six frontier models, with over 22% showing capability boundaries where scores swing by up to 90 points — related