Visual Question Answering on OpenWorldQA
70.5AccuracyPF-OPSD
Evaluation results
| Method | Links | |
|---|---|---|
| PF-OPSDProtocol=Qwen3.5-9B training baseline2026.06 | 70.5 | |
| GRPOProtocol=Qwen3.5-9B training baseline2026.06 | 61.2 | |
| SFTProtocol=Qwen3.5-9B training baseline2026.06 | 59.6 | |
| GPT-5.4Protocol=Zero-shot / no-simulation2026.06 | 53.4 | |
| Workflow Agent (Gemini-3-Flash)Protocol=Workflow-agent2026.06 | 49.5 | |
| Gemini-3-FlashProtocol=Zero-shot / no-simulation2026.06 | 48.2 | |
| Qwen3.6-27BProtocol=Zero-shot / no-simulation2026.06 | 41.4 | |
| Qwen3.5-9BProtocol=Zero-shot / no-simulation2026.06 | 39.8 | |
| Workflow Agent (Qwen3.5-9B)Protocol=Workflow-agent2026.06 | 38.6 | |
| HY3Protocol=Zero-shot / no-simulation2026.06 | 35 | |
| Qwen2.5-VL-7BProtocol=Zero-shot / no-simulation2026.06 | 14.2 |