Put the eggplant into the blue sink on AutoEval Real-world held-out tasks
72Success RateWorld-Gymnast
Evaluation Results
| Method | Links | |
|---|---|---|
| World-GymnastBackbone=600M parameter variant2026.02 | 72 | |
| Iter-SFTBackbone=OpenVLA 7B2026.02 | 10 | |
| SFTBackbone=OpenVLA 7B2026.02 | 4 |