GUI Action Prediction on Offline Benchmark
69.2Exact MatchClaude-4.5-Opus
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Claude-4.5-OpusGrounding=true2026.02 | 69.2 | 82.7 | |
| Gemini-3-FlashGrounding=true2026.02 | 69.2 | 82 | |
| UI-OceanusBackbone=Qwen2.5-VL-32B, Stage=CPT+SFT+GRPO2026.02 | 68.9 | 84.5 | |
| UI-OceanusBackbone=Qwen3-VL-32B, Stage=CPT+SFT2026.02 | 65 | 81.1 | |
| Claude-4.5-SonnetGrounding=true2026.02 | 64.8 | 81.2 | |
| UI-OceanusBackbone=Qwen2.5-VL-32B, Stage=CPT+SFT2026.02 | 64.8 | 82.1 | |
| Qwen3-VL-235B-A22BGrounding=true2026.02 | 61.5 | 79.3 | |
| Qwen3-VL-32BGrounding=true2026.02 | 61.2 | 79.3 | |
| GPT-5.2Grounding=true2026.02 | 61.1 | 76.9 | |
| Qwen2.5-VL-72BGrounding=true2026.02 | 56.8 | 77.2 | |
| Qwen2.5-VL-32BGrounding=true2026.02 | 56.5 | 77.9 | |
| Qwen2.5-VL-72BGrounding=false2026.02 | 51.4 | 77.2 | |
| Qwen3-VL-235B-A22BGrounding=false2026.02 | 47.5 | 79.4 | |
| Qwen3-VL-32BGrounding=false2026.02 | 47.1 | 79.2 | |
| Qwen2.5-VL-32BGrounding=false2026.02 | 33.9 | 56.6 |