Visual Question Answering on VDR
37.8AccuracyVision-DeepResearch-30B-A3B
Evaluation Results
| Method | Links | |
|---|---|---|
| Vision-DeepResearch-30B-A3BWorkflow=Agent Workflow2026.01 | 37.8 | |
| Vision-DeepResearch-8BWorkflow=Agent Workflow2026.01 | 29.2 | |
| Claude-3.7-SonnetWorkflow=Agent Workflow2026.01 | 27.2 | |
| Qwen3-VL-30B-A3B-ThinkingWorkflow=Agent Workflow2026.01 | 23.2 | |
| GPT-5Workflow=Agent Workflow2026.01 | 20.4 | |
| Qwen3-VL-30B-A3B-InstructWorkflow=Agent Workflow2026.01 | 20.2 | |
| Gemini-2.5 ProWorkflow=Agent Workflow2026.01 | 18.8 | |
| Qwen3-VL-8B-ThinkingWorkflow=Agent Workflow2026.01 | 17.6 | |
| Qwen3-VL-8B-InstructWorkflow=Agent Workflow2026.01 | 17 | |
| Gemini-2.5 FlashWorkflow=Agent Workflow2026.01 | 16.3 | |
| Claude-4-SonnetWorkflow=Agent Workflow2026.01 | 13.6 | |
| GPT-5Workflow=Direct Answer2026.01 | 9.8 | |
| Gemini-2.5 ProWorkflow=Direct Answer2026.01 | 8 | |
| Gemini-2.5 FlashWorkflow=Direct Answer2026.01 | 6.2 | |
| Qwen3-VL-8B-ThinkingWorkflow=Direct Answer2026.01 | 5.6 | |
| Claude-3.7-SonnetWorkflow=Direct Answer2026.01 | 4.6 | |
| Qwen3-VL-30B-A3B-ThinkingWorkflow=Direct Answer2026.01 | 4.4 | |
| Qwen3-VL-30B-A3B-InstructWorkflow=Direct Answer2026.01 | 3.8 | |
| Qwen3-VL-8B-InstructWorkflow=Direct Answer2026.01 | 2.8 | |
| Claude-4-SonnetWorkflow=Direct Answer2026.01 | 2 |