Educational Diagram Generation on 400 K-12 diagram prompts
97.2LEM ScoreGPT-4o
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| GPT-4oParadigm=Code-based (LLM→code)2026.04 | 97.2 | 0.8 | 184.5 | 2.1 | 0 | |
| Claude 3.5Paradigm=Code-based (LLM→code)2026.04 | 95.6 | 1.4 | 191.3 | 2 | 0 | |
| CAGEParadigm=Ours2026.04 | 92.4 | 2.6 | 97.1 | 3.9 | 0 | |
| GPT-4o imgParadigm=Closed-src. APIs2026.04 | 73.8 | 14.2 | 91.2 | 4.2 | 0.08 | |
| DALL·E 3Paradigm=Closed-src. APIs2026.04 | 64.3 | 19.7 | 98.4 | 4 | 0.04 | |
| GeminiParadigm=Closed-src. APIs2026.04 | 59.1 | 23.6 | 105.8 | 3.8 | 0.04 | |
| Flux.1-devParadigm=Open-src. diffusion2026.04 | 18.7 | 59.2 | 95.3 | 4.1 | 0 | |
| SD3 Med.Paradigm=Open-src. diffusion2026.04 | 14.9 | 64.8 | 103.7 | 3.9 | 0 | |
| SDXL 1.0Paradigm=Open-src. diffusion2026.04 | 11.3 | 71.4 | 112.6 | 3.8 | 0 |