Text-to-code generation on DTVBENCH
10.6Manim ScoreGPT-4o
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| GPT-4oModel Type=Proprietary2025.10 | 10.6 | 4.92 | |
| JANUSCODER-8BModel Type=Open-Source, Parameters=8B2025.10 | 9.7 | 6.07 | |
| Qwen2.5-Coder-32B-InsModel Type=Open-Source, Parameters=32B2025.10 | 9.61 | 4.98 | |
| Qwen2.5-Coder-7B-InsModel Type=Open-Source, Parameters=7B2025.10 | 8.56 | 4.04 | |
| JANUSCODER-14BModel Type=Open-Source, Parameters=14B2025.10 | 8.41 | 5.97 | |
| Qwen3-14BModel Type=Open-Source, Parameters=14B2025.10 | 6.63 | 5.08 | |
| Qwen3-8BModel Type=Open-Source, Parameters=8B2025.10 | 6.2 | 5.18 | |
| LLaMA3-8B-InstructModel Type=Open-Source, Parameters=8B2025.10 | 4.92 | 3.15 |