GUI Agent Planning on GUIAct-Web 2024a (test)
50.5Step Success RateGPT-4o + GoClick-L (Intent)
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| GPT-4o + GoClick-L (Intent)Planner=GPT-4o, Grounding Model=GoClick-L w/ Intent Gnd.2026.04 | 50.5 | 62 | |
| GPT-4o + GoClick-L (Func)Planner=GPT-4o, Grounding Model=GoClick-L w/ Func. Gnd.2026.04 | 47.8 | 57.2 | |
| GPT-4o + SoMPlanner=GPT-4o + SoM, Grounding Model=None2026.04 | 42.3 | 55.6 | |
| Gemini-2-Flash-Exp + GoClick-L (Intent)Planner=Gemini-2-Flash-Exp, Grounding Model=GoClick-L w/ Intent Gnd.2026.04 | 41.7 | 51.6 | |
| Gemini-2-Flash-Exp + GoClick-L (Func)Planner=Gemini-2-Flash-Exp, Grounding Model=GoClick-L w/ Func. Gnd.2026.04 | 39.9 | 48.5 | |
| Gemini-2-Flash-Exp + SoMPlanner=Gemini-2-Flash-Exp + SoM, Grounding Model=None2026.04 | 32.9 | 44.7 | |
| GPT-4oPlanner=GPT-4o, Grounding Model=None2026.04 | 18.2 | 5.1 | |
| Gemini-2-Flash-ExpPlanner=Gemini-2-Flash-Exp, Grounding Model=None2026.04 | 16.8 | 8 |