Visual Code Generation on ScreenBench
76.8Block ScoreUniCoder
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| UniCoderConfiguration=Full Model2026.06 | 76.8 | 85.7 | 75.5 | 73.4 | 81.2 | |
| UniCoderConfiguration=w/o Attribute Reward2026.06 | 76.2 | 85.2 | 68.5 | 68 | 80.1 | |
| Claude-4.5-OpusModel Category=Closed-Source Models2026.06 | 76.1 | 84.9 | 74.2 | 72.5 | 80.5 | |
| GPT-5Model Category=Closed-Source Models2026.06 | 75.8 | 85.4 | 72.8 | 71.8 | 80.2 | |
| Gemini-3-ProModel Category=Closed-Source Models2026.06 | 75.2 | 85.1 | 73.5 | 71.2 | 79.8 | |
| UniCoderConfiguration=w/o Reference Injection2026.06 | 74.8 | 84.6 | 71 | 70.5 | 79.2 | |
| Qwen3-VL-8BModel Category=Open-Source Models2026.06 | 74.1 | 84.3 | 65.2 | 66.1 | 78.4 | |
| Qwen2.5-VL-7BModel Category=Open-Source Models2026.06 | 72.3 | 82.8 | 61.3 | 63.2 | 76.2 | |
| InternVL3.5-8BModel Category=Open-Source Models2026.06 | 71 | 82.2 | 59.8 | 62 | 75.5 | |
| Deepseek-VL-7BModel Category=Open-Source Models2026.06 | 68 | 77.3 | 57 | 61.4 | 73.2 | |
| LLaVA-v1.6-7BModel Category=Open-Source Models2026.06 | 63.5 | 83 | 54.4 | 59.2 | 72.7 |