Computer task execution on OSWorld 361 tasks
54.65Overall Success RateAgentS3
Evaluation Results
| Method | Links | |||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| AgentS3Configuration=+ Plan. & Gnd. (GUIDE full dual-channel), Worker=GPT-5.2, Grounding=Seed-1.8 (Mode A), k=52026.03 | 54.65 | 49.85 | 53.85 | 65.96 | 46.88 | 65.22 | 70.83 | 80 | 56.25 | 82.61 | 37.1 | |
| Watch & LearnConfiguration=+Video (ICL)2026.03 | 52.8 | — | — | — | — | — | — | — | — | — | — | |
| Watch & LearnConfiguration=No Annotation2026.03 | 50.6 | — | — | — | — | — | — | — | — | — | — | |
| AgentS3Configuration=No Annotation, Worker=GPT-5.2, Grounding=Seed-1.8 (Mode A), k=52026.03 | 50.18 | 41.18 | 38.46 | 51.06 | 44.62 | 52.17 | 70.83 | 73.33 | 73.91 | 73.91 | 40.32 | |
| Seed-1.8Configuration=+ Plan. & Gnd. (GUIDE full dual-channel)2026.03 | 44.62 | 47.74 | 42.31 | 48.94 | 45.31 | 56.51 | 50 | 73.33 | 52.32 | 65.22 | 25.74 | |
| Seed-1.8Configuration=+ Planning (GUIDE)2026.03 | 43.93 | 47.74 | 34.62 | 42.55 | 46.51 | 56.38 | 50 | 73.33 | 52.32 | 65.22 | 27.89 | |
| Qwen3-VL-8BConfiguration=+ Plan. & Gnd. (GUIDE full dual-channel)2026.03 | 39.73 | 36.87 | 46.15 | 34.78 | 38.48 | 52.16 | 54.17 | 80 | 51.77 | 43.48 | 25.98 | |
| Qwen3-VL-8BConfiguration=+ Planning (GUIDE)2026.03 | 38.93 | 41.22 | 48 | 27.66 | 47.59 | 52.16 | 50 | 73.33 | 40.56 | 52.17 | 21.71 | |
| Seed-1.8Configuration=No Annotation (baseline)2026.03 | 37.14 | 36.87 | 26.92 | 29.79 | 43.09 | 34.77 | 45.83 | 66.67 | 47.06 | 60.87 | 26.88 | |
| Qwen3-VL-8BConfiguration=No Annotation (baseline)2026.03 | 33.9 | — | — | — | — | — | — | — | — | — | — |