Magma-Edu: Multi-Agent Generative Multimodal Framework For Text-Diagram Educational Question Generation

Zhenyu Wu, Beijing Normal University

Jian Li, Beijing Normal University

Hua Huang, Beijing Normal University

Date

11/2025

AI Overview

Why it matters

Educational diagrams help convey abstract concepts, but multimodal models struggle to generate pedagogically coherent problems with matching, geometrically accurate figures.

Study at a glance

Question: Can a multi-agent system improve the accuracy and text-diagram consistency of generated math problems?
Setting and data: 78 K-12 junior and senior high mathematics knowledge points, used to generate 390 candidate multimodal questions.
Method: Multi-agent framework evaluation and ablation experiments comparing baseline models against the iterative system across text and visual metrics.
AI tools: MAGMA-Edu, GPT-4o, GPT-5, DeepSeek-V3.1, DeepSeek-R1, Gemini 2.5 Pro

Key findings

Applying MAGMA-Edu to GPT-4o increased average text quality scores from 57.01 to 92.31 and image-text consistency from 13.20 to 85.24.
Across all tested models, Gemini 2.5 Pro with MAGMA-Edu achieved the highest scores, reaching 96.20 in text quality and 99.12 in image-text consistency.
Ablation results showed that Stage 1 drove textual improvements, while Stage 2's code-based rendering drove large gains in visual consistency across all models.

Summary drafted with AI from the full paper and may contain errors. See the full paper for details.

Educational illustrations play a central role in communicating abstract concepts, yet current multimodal large language models (MLLMs) remain limited in producing pedagogically coherent and semantically consistent educational visuals. We introduce MAGMA-Edu, a self-reflective multi-agent framework that unifies textual reasoning and diagrammatic synthesis for structured educational problem generation. Unlike existing methods that treat text and image generation independently, MAGMA-Edu employs a two-stage co-evolutionary pipeline: (1) a generation-verification-reflection loop that iteratively refines question statements and solutions for mathematical accuracy, and (2) a code-based intermediate representation that enforces geometric fidelity and semantic alignment during image rendering. Both stages are guided by internal self-reflection modules that evaluate and revise outputs until domain-specific pedagogical constraints are met. Extensive experiments on multimodal educational benchmarks demonstrate the superiority of MAGMA-Edu over state-of-the-art MLLMs. Compared to GPT-4o, MAGMA-Edu improves the average textual metric from 57.01 to 92.31 (+35.3 pp) and boosts image-text consistency (ITC) from 13.20 to 85.24 (+72 pp). Across all model backbones, MAGMA-Edu achieves the highest scores (Avg-Text 96.20, ITC 99.12), establishing a new state of the art for multimodal educational content generation and demonstrating the effectiveness of self-reflective multi-agent collaboration in pedagogically aligned vision-language reasoning.