Multimodal Research on In-house 1
52.5AccuracyGemini-3-Flash
Evaluation Results
| Method | Links | |
|---|---|---|
| Gemini-3-FlashAgent Type=Direct Answer2026.04 | 52.5 | |
| GPT-5.4Agent Type=Direct Answer2026.04 | 45.1 | |
| MIAAgent Type=Memory-based Search Agent2026.04 | 31.8 | |
| Gemini-2.5-ProAgent Type=Direct Answer2026.04 | 30.8 | |
| Unsupervised MIAAgent Type=Memory-based Search Agent2026.04 | 29.8 | |
| Qwen2.5-VL-32B+ReACTAgent Type=Search Agent2026.04 | 28.8 | |
| GPT-4oAgent Type=Direct Answer2026.04 | 25.6 | |
| MementoAgent Type=Memory-based Search Agent2026.04 | 22.7 | |
| ExpeLAgent Type=Memory-based Search Agent2026.04 | 19.7 | |
| Qwen2.5-VL-32BAgent Type=Direct Answer2026.04 | 18.6 | |
| ReasoningBankAgent Type=Memory-based Search Agent2026.04 | 18.6 | |
| No MemoryAgent Type=Memory-based Search Agent2026.04 | 15.9 | |
| MMSearch-R1Agent Type=Search Agent2026.04 | 13.6 | |
| RAGAgent Type=Memory-based Search Agent2026.04 | 12.5 | |
| Mem0Agent Type=Memory-based Search Agent2026.04 | 12.5 | |
| A-MemAgent Type=Memory-based Search Agent2026.04 | 12.5 | |
| Qwen2.5-VL-7BAgent Type=Direct Answer2026.04 | 9.5 | |
| Qwen2.5-VL-7B+ReACTAgent Type=Search Agent2026.04 | 9.5 |