Scientific Graphics Program Synthesis on Scientific Graphics Program Synthesis Benchmark (test)
7.12Visual Fidelity (VF)SciTikZer-8B
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| SciTikZer-8BParameters=8B2026.04 | 7.12 | 8.05 | 8.4 | 7.86 | 23.57 | |
| Gemini-2.5-ProModel tier=Pro2026.04 | 7.03 | 8.01 | 8.29 | 7.78 | 23.33 | |
| GPT-5.12026.04 | 6.41 | 7.43 | 8.09 | 7.31 | 21.93 | |
| DeTikZify-v2.5-8BParameters=8B, Version=v2.52026.04 | 6.35 | 7.48 | 7.45 | 7.09 | 21.28 | |
| Qwen3-VL-Instruct-32BParameters=32B2026.04 | 5.82 | 6.76 | 7.02 | 6.53 | 19.6 |