Close the drawer on AutoEval Real-world held-out tasks
62Success RateSFT
Evaluation Results
| Method | Links | |
|---|---|---|
| SFTBackbone=OpenVLA 7B2026.02 | 62 | |
| World-GymnastBackbone=600M parameter variant2026.02 | 62 | |
| Iter-SFTBackbone=OpenVLA 7B2026.02 | 60 |
| Method | Links | |
|---|---|---|
| SFTBackbone=OpenVLA 7B2026.02 | 62 | |
| World-GymnastBackbone=600M parameter variant2026.02 | 62 | |
| Iter-SFTBackbone=OpenVLA 7B2026.02 | 60 |