Web Navigation on Mind2Web Cross-Domain (EA, AF1, StepSR, TaskSR)
68.4Element Accuracy (EA)GUICrafter-7B
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| GUICrafter-7BModel Size=About 7B2026.06 | 68.4 | 91.3 | 62.2 | — | |
| GUICrafter-7B (only Stage2)Model Size=About 7B, Training Stage=Stage 22026.06 | 66.7 | 89.7 | 60.1 | — | |
| UI-TARS-7BModel Size=About 7B2026.06 | 66.6 | 90.9 | 60.5 | — | |
| GUI-R1-7B (v2)Model Size=About 7B2026.06 | 65.9 | 88.6 | 59 | — | |
| FluxMemBase Model=GPT-4.1-mini, Setting=Filtered (manual removal of non-golden HTML elements)2026.05 | 65.2 | 75.3 | 55.3 | 7.2 | |
| FluxMemBase Model=Gemini-2.5-flash, Setting=Filtered (manual removal of non-golden HTML elements)2026.05 | 64.8 | 70.3 | 54.6 | 7 | |
| GPT-4o + Aria-UIAgent Framework=Aria-UI [47]2026.06 | 61.4 | — | — | — | |
| GUICrafter-3BModel Size=About 3B2026.06 | 60.7 | 90.6 | 52.9 | — | |
| Aguvis-7BModel Size=About 7B2026.06 | 60.4 | 89.2 | 56.6 | — | |
| Qwen2.5-VL-7B (Fine-tuned)Model Size=About 7B, Protocol=Fine-tuned2026.06 | 60.2 | 89.2 | 56.4 | — | |
| UI-TARS-2BModel Size=About 3B2026.06 | 58.8 | 89.6 | 52.3 | — | |
| FluxMemBase Model=Gemini-2.5-flash, Setting=Realistic (without any pre-filtering)2026.05 | 58.5 | 70.1 | 55.9 | 5 | |
| FluxMemBase Model=GPT-4.1-mini, Setting=Realistic (without any pre-filtering)2026.05 | 56.2 | 69.4 | 52.1 | 4.3 | |
| GUICrafter-3B (only Stage2)Model Size=About 3B, Training Stage=Stage 22026.06 | 56.2 | 86.7 | 50.7 | — | |
| GUI-R1-3B (v2)Model Size=About 3B2026.06 | 54.7 | 85.8 | 49.8 | — | |
| Upper BoundBackbone=Qwen2.5VL-3B2026.03 | 53.25 | 60.27 | 48.89 | — | |
| DivPruneBackbone=Qwen2.5VL-3B2026.03 | 52.89 | 58.96 | 48.54 | — | |
| PDropBackbone=Qwen2.5VL-3B2026.03 | 52.81 | 59.98 | 48.42 | — | |
| RandomBackbone=Qwen2.5VL-3B2026.03 | 52.65 | 58.78 | 48.13 | — | |
| FASTVBackbone=Qwen2.5VL-3B2026.03 | 52.55 | 58.3 | 48.28 | — | |
| SparseVLMBackbone=Qwen2.5VL-3B2026.03 | 52.44 | 58.47 | 48.16 | — | |
| AWMBase Model=Gemini-2.5-flash, Setting=Filtered (manual removal of non-golden HTML elements)2026.05 | 52 | 67 | 47.9 | 3.4 | |
| DartBackbone=Qwen2.5VL-3B2026.03 | 51.97 | 59.33 | 47.75 | — | |
| GUICrafter-7B (only Stage1)Model Size=About 7B, Training Stage=Stage 12026.06 | 51 | 58.8 | 46.1 | — | |
| AWMBase Model=GPT-4.1-mini, Setting=Filtered (manual removal of non-golden HTML elements)2026.05 | 50.5 | 69.3 | 43.9 | 2 | |
| Upper BoundBackbone=Qwen2VL-2B2026.03 | 47.6 | 57.86 | 43.99 | — | |
| FASTVBackbone=Qwen2VL-2B2026.03 | 47.34 | 57.21 | 43.71 | — | |
| PDropBackbone=Qwen2VL-2B2026.03 | 47.2 | 58.07 | 43.66 | — | |
| Qwen2.5-VL-3B (Fine-tuned)Model Size=About 3B, Protocol=Fine-tuned2026.06 | 47 | 55.7 | 43.2 | — | |
| RandomBackbone=Qwen2VL-2B2026.03 | 46.7 | 57.09 | 43.32 | — | |
| DivPruneBackbone=Qwen2VL-2B2026.03 | 46.66 | 57.88 | 43.26 | — | |
| DartBackbone=Qwen2VL-2B2026.03 | 46.62 | 57.25 | 43.04 | — | |
| GPT-4o + UGroundAgent Framework=UGround [10]2026.06 | 46.6 | — | — | — | |
| GPT-4V + OmniParserAgent Framework=OmniParser [22]2026.06 | 45.5 | 85.7 | 42 | — | |
| SparseVLMBackbone=Qwen2VL-2B2026.03 | 45.28 | 57 | 41.68 | — | |
| GUI-R1-7B (v1)Model Size=About 7B2026.06 | 44.6 | 47 | 35.7 | — | |
| ReasoningBankBackbone=Gemini-2.5-pro2025.09 | 42.8 | 45.2 | 38.1 | 1.7 | |
| UI-R1-3BModel Size=About 3B2026.06 | 42.5 | — | — | — | |
| HMTBackbone=GPT-42026.03 | 42.3 | 48.8 | 37.1 | 1.9 | |
| AWMBase Model=Gemini-2.5-flash, Setting=Realistic (without any pre-filtering)2026.05 | 41.6 | 60.2 | 38.1 | 1.5 | |
| AWMSetting=online2026.03 | 40.9 | 46.3 | 35.5 | 1.7 | |
| ReasoningBankBackbone=Gemini-2.5-flash2025.09 | 40.6 | 41.3 | 36.6 | 1.6 | |
| ReasoningBankBase Model=Gemini-2.5-flash, Setting=Filtered (manual removal of non-golden HTML elements)2026.05 | 40.6 | 41.3 | 36.6 | 1.6 | |
| GUICrafter-3B (only Stage1)Model Size=About 3B, Training Stage=Stage 12026.06 | 40.4 | 37.5 | 36.7 | — | |
| ShowUI-2B (Fine-tuned)Model Size=About 3B, Protocol=Fine-tuned2026.06 | 39.4 | 86.8 | 35.2 | — | |
| GUI-R1-3B (v1)Model Size=About 3B2026.06 | 38.9 | 35.9 | 32 | — | |
| SynapseBackbone=Gemini-2.5-pro2025.09 | 38.5 | 39.8 | 35.6 | 1.5 | |
| No MemoryBackbone=Gemini-2.5-pro2025.09 | 37.9 | 37.7 | 35 | 1.4 | |
| AWMBackbone=Gemini-2.5-pro2025.09 | 37.3 | 38.1 | 34.4 | 1.2 | |
| AWMSetting=offline2026.03 | 36.4 | 41.6 | 32.6 | 0.7 | |
| Qwen2.5-VL-7B (Zero-Shot)Model Size=About 7B, Protocol=Zero-Shot2026.06 | 36.4 | 40.4 | 25.3 | — | |
| SynapseBackbone=Gemini-2.5-flash2025.09 | 36.3 | 38.5 | 32.4 | 1.1 | |
| No MemoryBackbone=Gemini-2.5-flash2025.09 | 35.8 | 37.9 | 31.9 | 1 | |
| GPT-4 (Text-only)Input Modality=Text-only2026.06 | 35.4 | 61.9 | 29.7 | — | |
| GPT-4o + SeeClickAgent Framework=SeeClick [6]2026.06 | 33.5 | — | — | — | |
| AWMBackbone=Gemini-2.5-flash2025.09 | 33.3 | 36.5 | 30.1 | 0.7 | |
| AWMBase Model=GPT-4.1-mini, Setting=Realistic (without any pre-filtering)2026.05 | 30.3 | 56.9 | 24.9 | 1 | |
| No MemoryBase Model=GPT-4.1-mini, Setting=Realistic (without any pre-filtering)2026.05 | 30.1 | 55.9 | 24.6 | 0.7 | |
| Qwen2.5-VL-3B (Zero-Shot)Model Size=About 3B, Protocol=Zero-Shot2026.06 | 28.8 | 34.1 | 22.6 | — | |
| GPT-4 (SOM)Input Modality=SOM2026.06 | 27 | — | 23.7 | — | |
| ShowUI-2B (Zero-Shot)Model Size=About 3B, Protocol=Zero-Shot2026.06 | 24.4 | 83.9 | 21.4 | — | |
| No MemoryBase Model=Gemini-2.5-flash, Setting=Realistic (without any pre-filtering)2026.05 | 22.1 | 37.2 | 18.6 | 1.8 | |
| MindAct2026.03 | 21.6 | 52.8 | 26.4 | 2 | |
| GPT-4oCategory=Proprietary Model2026.06 | 5.5 | 86.4 | 4.5 | — |