Knowledge-intensive video QA on VideoSimpleQA
91.3AccuracyQwen-2.5VL-Instruct-32B
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Qwen-2.5VL-Instruct-32BRAG Strategy=Standard RAG, Model size=32B2026.01 | 91.3 | 47.72 | 69.03 | 44.07 | |
| VideoSpeculateRAGRAG Strategy=Speculate RAG, Model size=3B+32B2026.01 | 91.12 | 25.74 | 69.38 | 21.18 | |
| VideoSpeculateRAGRAG Strategy=Speculate RAG, Model size=3B+7B2026.01 | 82.6 | 23.04 | 61.8 | 19.54 | |
| InternVL-3-38BRAG Strategy=Standard RAG, Model size=38B2026.01 | 78.86 | 14 | 61.18 | 10.88 | |
| LLaVA-NeXT-Video-34BRAG Strategy=Standard RAG, Model size=34B2026.01 | 76.8 | 56.54 | 57.15 | 53.96 | |
| Qwen-2.5VL-Instruct-3BRAG Strategy=Standard RAG, Model size=3B2026.01 | 75.26 | 4.84 | 57.13 | 4.8 | |
| Gemini-3-Pro2026.06 | 71.9 | — | — | — | |
| Seed2.0 Pro2026.06 | 71.9 | — | — | — | |
| Gemini-3-Flash2026.06 | 70.7 | — | — | — | |
| Seed1.82026.06 | 67.8 | — | — | — | |
| Seed2.0 Mini2026.06 | 67.7 | — | — | — | |
| Seed2.0 Lite2026.06 | 66.6 | — | — | — | |
| Qwen-2.5VL-Instruct-32BRAG Strategy=No RAG, Model size=32B2026.01 | 57.73 | 19.18 | 39.87 | 19.65 | |
| InternVL-3-38BRAG Strategy=No RAG, Model size=38B2026.01 | 40.2 | 4.06 | 26.85 | 4.84 | |
| Qwen-2.5VL-Instruct-3BRAG Strategy=No RAG, Model size=3B2026.01 | 37.62 | 2.15 | 25.81 | 2.72 | |
| LLaVA-NeXT-Video-34BRAG Strategy=No RAG, Model size=34B2026.01 | 27.31 | 29.24 | 17.92 | 27.46 |