Multimodal Search on MMSearch
74.27AccuracyGemini-3-Pro
Evaluation results
| Method | Links | |
|---|---|---|
| Gemini-3-ProWorkflow=Agentic Model (zero-shot), Model Type=Proprietary2026.06 | 74.27 | |
| Qwen3-VL-30B + ODE-RLEvaluation Setting=Visual-Native Agent Harness, Training Strategy=ODE-RL2026.05 | 69.7 | |
| MM-DeepResearch 32BEvaluation Paradigm=Agentic Search2026.03 | 69 | |
| Gemini-3-FlashAgent Type=Direct Answer2026.04 | 69 | |
| MM-DeepResearch-8BEvaluation Paradigm=Agentic Search2026.03 | 67.8 | |
| SenseNova-MARS-8BEvaluation Paradigm=Agentic Search2026.03 | 67.4 | |
| TAPOWorkflow=Agentic Model, Model Type=Open-source, Backbone=SAPO2026.06 | 67.21 | |
| Claude-4-SonnetEvaluation Setting=Agent Workflow2026.05 | 67.2 | |
| SenseNova-MARSWorkflow=Agentic Model, Model Type=Open-source2026.06 | 66.67 | |
| SkyWork-R1V4Workflow=Agentic Model, Model Type=Open-source2026.06 | 66.1 | |
| GPT-5.2Workflow=Agentic Model (zero-shot), Model Type=Proprietary2026.06 | 66.08 | |
| Qwen3-VL-8B + ODE-RLEvaluation Setting=Visual-Native Agent Harness, Training Strategy=ODE-RL2026.05 | 66 | |
| Qwen3-VL-30B + ODE-SFTEvaluation Setting=Visual-Native Agent Harness, Training Strategy=ODE-SFT2026.05 | 65.7 | |
| Qwen3-VL-8B-Instruct+SAPOWorkflow=Agentic Model, Model Type=Open-source2026.06 | 64.33 | |
| DeepEyes-v2-7BEvaluation Paradigm=Agentic Search2026.03 | 63.7 | |
| Deepeyes2Agent Type=Search Agent2026.04 | 63.7 | |
| Deepeyes2Evaluation Paradigm=Search Agent2026.04 | 63.7 | |
| Claude-3.7-SonnetEvaluation Setting=Agent Workflow2026.05 | 63.7 | |
| DeepEyesV2Workflow=Agentic Model, Model Type=Open-source2026.06 | 63.7 | |
| GPT-5Evaluation Setting=Agent Workflow2026.05 | 62.7 | |
| MIAAgent Type=Memory-based Search Agent2026.04 | 62.6 | |
| Gemini-3-ProWorkflow=Direct Answer, Model Type=Proprietary2026.06 | 62.57 | |
| Gemini-3-FlashWorkflow=Agentic Model (zero-shot), Model Type=Proprietary2026.06 | 62.57 | |
| Qwen3-VL-8B + ODE-SFTEvaluation Setting=Visual-Native Agent Harness, Training Strategy=ODE-SFT2026.05 | 61.7 | |
| MM-DeepResearch-7BEvaluation Paradigm=Agentic Search2026.03 | 61.4 | |
| ExpeLAgent Type=Memory-based Search Agent2026.04 | 61.4 | |
| MementoAgent Type=Memory-based Search Agent2026.04 | 61.4 | |
| DR-MMSearchAgentEvaluation Paradigm=Search Agent2026.04 | 61 | |
| Unsupervised MIAAgent Type=Memory-based Search Agent2026.04 | 60.2 | |
| Gemini-2.5-FlashWorkflow=Agentic Model (zero-shot), Model Type=Proprietary2026.06 | 59.06 | |
| Gemini-3-ProModel Type=Proprietary, Tool Usage=Web Tools2026.03 | 59 | |
| Gemini-3-FlashWorkflow=Direct Answer, Model Type=Proprietary2026.06 | 57.31 | |
| Gemini-3-FlashModel Type=Proprietary, Tool Usage=Web Tools2026.03 | 57.3 | |
| ReasoningBankAgent Type=Memory-based Search Agent2026.04 | 57.3 | |
| Qwen3-VL-30BEvaluation Setting=Visual-Native Agent Harness2026.05 | 57 | |
| Gemini-2.5 ProEvaluation Setting=Agent Workflow2026.05 | 55.7 | |
| No MemoryAgent Type=Memory-based Search Agent2026.04 | 55.6 | |
| WebWatcher-32BEvaluation Paradigm=Agentic Search2026.03 | 55.3 | |
| WebWatcherAgent Type=Search Agent2026.04 | 55.3 | |
| WebWatcherEvaluation Paradigm=Search Agent2026.04 | 55.3 | |
| WebWatcher-32BEvaluation Setting=Deep Search Agent2026.05 | 55.3 | |
| RAGAgent Type=Memory-based Search Agent2026.04 | 54.4 | |
| MMSearch-R1-7BEvaluation Paradigm=Agentic Search2026.03 | 53.8 | |
| MMSearch-R1Evaluation Paradigm=Search Agent2026.04 | 53.8 | |
| MMSearch-R1-7BEvaluation Setting=Deep Search Agent2026.05 | 53.8 | |
| MMSearch-R1Workflow=Agentic Model, Model Type=Open-source2026.06 | 53.8 | |
| Gemini-2.5 FlashEvaluation Setting=Agent Workflow2026.05 | 53.3 | |
| GPT-5Workflow=Agentic Model (zero-shot), Model Type=Proprietary2026.06 | 52.63 | |
| GPT-5Evaluation Paradigm=RAG Workflow2026.03 | 52.6 | |
| Qwen3-VL-PlusModel Type=Proprietary, Tool Usage=Web Tools2026.03 | 50.7 | |
| Qwen2.5-VL-32B-InstructWorkflow=Agentic Model (zero-shot), Model Type=Open-source2026.06 | 49.71 | |
| Qwen3-VL-30BEvaluation Setting=Agent Workflow2026.05 | 49.3 | |
| Qwen3-VL-32B-InstructWorkflow=Agentic Model (zero-shot), Model Type=Open-source2026.06 | 49.12 | |
| GPT-4oWorkflow=Agentic Model (zero-shot), Model Type=Proprietary2026.06 | 49.12 | |
| GPT-4oEvaluation Paradigm=RAG Workflow2026.03 | 49.1 | |
| WebWatcher-7BEvaluation Paradigm=Agentic Search2026.03 | 49.1 | |
| WebWatcher-7BEvaluation Setting=Deep Search Agent2026.05 | 49.1 | |
| Qwen3-VL-8BEvaluation Setting=Agent Workflow2026.05 | 48.7 | |
| Qwen3-VL-8B-InstructWorkflow=Agentic Model (zero-shot), Model Type=Open-source2026.06 | 47.37 | |
| Qwen3-VL-8BEvaluation Paradigm=RAG Workflow2026.03 | 47.3 | |
| Qwen3-VL-8BEvaluation Setting=Visual-Native Agent Harness2026.05 | 47.3 | |
| VSearcherModel Type=Agentic2026.03 | 47.2 | |
| GPT-5Model Type=Proprietary, Tool Usage=Web Tools2026.03 | 45.2 | |
| Qwen3-VL-32BEvaluation Paradigm=Agentic Search2026.03 | 44.4 | |
| GPT-5.4Agent Type=Direct Answer2026.04 | 44.4 | |
| MMSearch-R1Agent Type=Search Agent2026.04 | 43.9 | |
| Qwen3-VL-30B-A3B-ThinkingModel Type=Open-source, Tool Usage=Web Tools2026.03 | 43.7 | |
| Mem0Agent Type=Memory-based Search Agent2026.04 | 43.3 | |
| GPT-5.2Workflow=Direct Answer, Model Type=Proprietary2026.06 | 43.27 | |
| Qwen2.5-VL-72B-InstructModel Type=Open-source, Tool Usage=Web Tools2026.03 | 41.5 | |
| DeepEyesV2Model Type=Agentic2026.03 | 40.9 | |
| A-MemAgent Type=Memory-based Search Agent2026.04 | 40.9 | |
| GPT-5Evaluation Setting=Direct Reasoning2026.05 | 40.7 | |
| Gemini-2.5 ProEvaluation Setting=Direct Reasoning2026.05 | 39.8 | |
| Qwen3-VL-8B-ThinkingModel Type=Open-source, Tool Usage=Web Tools2026.03 | 39.1 | |
| GPT-4o-miniWorkflow=Agentic Model (zero-shot), Model Type=Proprietary2026.06 | 38.6 | |
| MMSearch-R1Model Type=Agentic2026.03 | 37.5 | |
| Qwen3-VL-8BEvaluation Paradigm=Agentic Search2026.03 | 37.4 | |
| GPT-5Evaluation Paradigm=Direct Reasoning2026.03 | 35.1 | |
| GPT-5Workflow=Direct Answer, Model Type=Proprietary2026.06 | 35.09 | |
| Qwen3-VL-4B-ThinkingModel Type=Open-source, Tool Usage=Web Tools2026.03 | 35 | |
| Visual-ARFT-7BEvaluation Paradigm=Agentic Search2026.03 | 34.5 | |
| Visual-ARFTWorkflow=Agentic Model, Model Type=Open-source2026.06 | 34.5 | |
| InternVL3.5-38BModel Type=Open-source, Tool Usage=Web Tools2026.03 | 33.3 | |
| Qwen2.5-VL-7B-InstructWorkflow=Agentic Model (zero-shot), Model Type=Open-source2026.06 | 32.16 | |
| Gemini-2.5 FlashEvaluation Setting=Direct Reasoning2026.05 | 30.4 | |
| InternVL3.5-8BModel Type=Open-source, Tool Usage=Web Tools2026.03 | 29.2 | |
| Qwen2.5-VL-32B+ReACTAgent Type=Search Agent2026.04 | 27.3 | |
| Qwen2.5-VL-32BEvaluation Paradigm=Search Agent2026.04 | 27.3 | |
| Gemini-2.5-ProAgent Type=Direct Answer2026.04 | 26.9 | |
| Gemini2.5 ProEvaluation Paradigm=Direct2026.04 | 26.9 | |
| GPT-4oWorkflow=Direct Answer, Model Type=Proprietary2026.06 | 23.39 | |
| PIXELRAGReader=Qwen3-VL-4B, top-k=1, Retrieval Modality=Pixel-based retrieval2026.06 | 22.7 | |
| GPT-4oAgent Type=Direct Answer2026.04 | 22.2 | |
| GPT-4oEvaluation Paradigm=Direct2026.04 | 22.2 | |
| Gemini-2.5-FlashWorkflow=Direct Answer, Model Type=Proprietary2026.06 | 21.64 | |
| Qwen2.5-VL-7B+ReACTAgent Type=Search Agent2026.04 | 21.1 | |
| Qwen2.5-VL-7BEvaluation Paradigm=Search Agent2026.04 | 21.1 | |
| Claude-3.7-SonnetEvaluation Setting=Direct Reasoning2026.05 | 21.1 | |
| PIXELRAG (base)Reader=Qwen3-VL-4B, top-k=1, Retrieval Modality=Pixel-based retrieval2026.06 | 20.3 |