Open the drawer on AutoEval Real-world held-out tasks
0.58Success RateWorld-Gymnast
Evaluation results
| Method | Links | |
|---|---|---|
| World-GymnastBackbone=600M parameter variant, Pretraining Dataset=Open X-Embodiment2026.02 | 0.58 | |
| SFTBackbone=OpenVLA 7B, Fine-tuning Dataset=BridgeData V22026.02 | 0.4 | |
| Iter-SFTBackbone=OpenVLA 7B, Sampling rate=1:12026.02 | 0.3 |