Multi-step Visual Generation on VisGenBench
99Success Rate (Image)VisionCreator-32B
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| VisionCreator-32BSystem Prompt=New version2026.03 | 99 | 96 | 3.53 | 3.49 | 3.42 | |
| Qwen3-VL-32B-ThinkingSystem Prompt=New version2026.03 | 97 | 93 | 3.47 | 3.23 | 3.18 | |
| Qwen3-VL-32B-SFTSystem Prompt=New version2026.03 | 96 | 94 | 3.53 | 3.37 | 3.27 | |
| GPT-5System Prompt=New version2026.03 | 95.95 | 93 | 3.52 | 3.25 | 3.19 | |
| Gemini2.5-ProSystem Prompt=New version2026.03 | 91 | 84 | 3.53 | 3.35 | 3.01 | |
| Qwen3-VL-32B-RLSystem Prompt=New version2026.03 | 91 | 87 | 3.51 | 3.4 | 3.07 |