Deep Research Report Generation on MMR Bench+
4.15InformativenessViDR (GPT-5.2)
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| ViDR (GPT-5.2)Source Access Type=Open Source, Backbone Model=GPT-5.22026.05 | 4.15 | 4 | 4.03 | 3.85 | 3.66 | 3.94 | |
| ViDR (Gemini-3-Flash)Source Access Type=Open Source, Backbone Model=Gemini-3-Flash2026.05 | 4.09 | 3.66 | 3.53 | 3.97 | 3.42 | 3.74 | |
| OpenAI-DRSource Access Type=Closed Source2026.05 | 4.03 | 3.86 | 3.48 | 1.71 | 1 | 2.82 | |
| Qwen-Deep-ResearchSource Access Type=Closed Source2026.05 | 3.98 | 3.93 | 2.84 | 1.82 | 1 | 2.71 | |
| Gemini-DRSource Access Type=Closed Source2026.05 | 3.97 | 3.71 | 3.09 | 1.72 | 1.02 | 2.7 | |
| Baidu Qianfan DR-agentSource Access Type=Closed Source2026.05 | 3.82 | 3.87 | 2.73 | 1.34 | 1 | 2.55 | |
| MMDR(GPT-5.2)Source Access Type=Open Source, Backbone Model=GPT-5.22026.05 | 3.44 | 3.89 | 3.03 | 3.62 | 1 | 3 | |
| Tongyi-DR(30B)Source Access Type=Open Source, Backbone Model=30B2026.05 | 3.31 | 3.64 | 2.29 | 1.6 | 1.04 | 2.38 | |
| WebShaperSource Access Type=Open Source2026.05 | 3.13 | 3.5 | 2.75 | 1.23 | 1.05 | 2.33 |