EduIllustrate: Towards Scalable Automated Generation Of Multimodal Educational Content

2026-04-06

AI Summary

This study addresses the lack of systematic evaluation of large language models’ (LLMs’) capacity to generate multimodal instructional content for K–12 STEM education. The authors construct a benchmark comprising 230 items spanning five disciplines and three grade bands, and propose the first standardized evaluation framework for educational multimodal content generation. Their approach employs a sequential anchoring protocol to ensure geometric consistency between text and images as well as coherent reasoning, and introduces an eight-dimensional scoring rubric grounded in multimedia learning theory. High reliability on objective dimensions is demonstrated through validation by both human experts and LLM-as-judge. Experiments reveal that Gemini 3.0 Pro Preview achieves the highest performance (87.8%), while Kimi-K2.5 offers the best cost efficiency (80.8% at $0.12 per item). Sequential anchoring improves visual consistency by 13% and reduces generation costs by 94%.

Abstract

Large language models are increasingly used as educational assistants, yet evaluation of their educational capabilities remains concentrated on question-answering and tutoring tasks. A critical gap exists for multimedia instructional content generation -- the ability to produce coherent, diagram-rich explanations that combine geometrically accurate visuals with step-by-step reasoning. We present EduIllustrate, a benchmark for evaluating LLMs on interleaved text-diagram explanation generation for K-12 STEM problems. The benchmark comprises 230 problems spanning five subjects and three grade levels, a standardized generation protocol with sequential anchoring to enforce cross-diagram visual consistency, and an 8-dimension evaluation rubric grounded in multimedia learning theory covering both text and visual quality. Evaluation of ten LLMs reveals a wide performance spread: Gemini 3.0 Pro Preview leads at 87.8\%, while Kimi-K2.5 achieves the best cost-efficiency (80.8\% at \\$0.12/problem). Workflow ablation confirms sequential anchoring improves Visual Consistency by 13\% at 94\% lower cost. Human evaluation with 20 expert raters validates LLM-as-judge reliability for objective dimensions ($ρ\geq 0.83$) while revealing limitations on subjective visual assessment.

Problem

• multimodal educational content
• diagram-rich explanations
• LLM evaluation
• text-diagram generation
• K-12 STEM

Innovation

• multimodal educational content
• sequential anchoring
• visual consistency
• LLM evaluation benchmark
• diagram-rich explanation