GUI Agent Task on MemGUI-Bench Easy
41.7P@1Agent-S2
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Agent-S2Memory Type=Memory Agent, Backbone=Gemini-2.5-Pro2026.06 | 41.7 | 64.6 | 38.7 | |
| MemGUI-Agent-235BMemory Type=ConAct, Protocol=zero-shot, Backbone=Qwen3-VL-235B-Thinking2026.06 | 41.7 | 68.8 | 40.8 | |
| M3AMemory Type=Memory Agent, Backbone=Gemini-2.5-Pro2026.06 | 39.6 | 47.9 | 31.7 | |
| Qwen3-VL-32B-InstructMemory Type=Action-Thought2026.06 | 35.4 | 47.9 | 39.7 | |
| T3AMemory Type=Memory Agent, Backbone=Gemini-2.5-Pro2026.06 | 31.2 | 45.8 | — | |
| Qwen3-VL-235B-InstructMemory Type=Action-Thought2026.06 | 27.1 | 39.6 | 26.5 | |
| MemGUI-8B-SFTMemory Type=ConAct, Protocol=fine-tuned, Backbone=Qwen3-VL-8B-Instruct2026.06 | 25 | 39.6 | 20.5 | |
| GUI-Owl-1.5-32B-InstructMemory Type=Action-Thought2026.06 | 22.9 | 37.5 | 18.4 | |
| GUI-Owl-1.5-8B-InstructMemory Type=Action-Thought2026.06 | 22.9 | 31.2 | 15.7 | |
| Qwen3-VL-235B-ThinkingMemory Type=Action-Thought2026.06 | 18.8 | 47.9 | 21.8 | |
| Qwen3-VL-8B-InstructMemory Type=Action-Thought2026.06 | 18.8 | 35.4 | 20.1 | |
| GUI-Owl-7BMemory Type=Action-Thought2026.06 | 14.6 | 22.9 | 9.9 | |
| UI-Venus-7BMemory Type=Action-Thought2026.06 | 14.6 | 20.8 | 6.9 | |
| Mobile-Agent-EMemory Type=Memory Agent, Backbone=Gemini-2.5-Pro2026.06 | 12.5 | 22.9 | 0.9 | |
| Mobile-Agent-V2Memory Type=Memory Agent, Backbone=Gemini-2.5-Pro2026.06 | 8.3 | 10.4 | 0 | |
| AppAgentMemory Type=Action-Thought, Backbone=Gemini-2.5-Pro2026.06 | 8.3 | 22.9 | 4.6 | |
| UI-TARS-1.5-7BMemory Type=Multi-turn2026.06 | 8.3 | 16.7 | 9.3 | |
| SeeActMemory Type=Rule-based, Backbone=Gemini-2.5-Pro2026.06 | 6.2 | 12.5 | 0 | |
| CogAgentMemory Type=No History2026.06 | 0 | 0 | 0 |