LIBERO-Long (also called LIBERO-10) is one of four task suites in the LIBERO benchmark for lifelong robot learning. It contains 10 long-horizon manipulation tasks requiring multi-step reasoning and diverse object/spatial/goal knowledge. Reported as success rate (%).
5 results indexed across 1 metric. Shaded row marks current SOTA; ties broken by submission date.
| # | Model | Org | Submitted | Paper / code | success-rate |
|---|---|---|---|---|---|
| 01 | MolmoAct2-Think | — | May 2026 | MolmoAct2: Action Reasoning Models for Real-world Deploy… · code | 98.10 |
| 02 | MolmoAct2 | — | May 2026 | MolmoAct2: Action Reasoning Models for Real-world Deploy… · code | 97.20 |
| 03 | UD-VLA | — | Nov 2025 | Unified Diffusion VLA: Vision-Language-Action Model via … · code | 92.70 |
| 04 | SmolVLA (2.25B) | — | Jun 2025 | SmolVLA: A Vision-Language-Action Model for Affordable a… · code | 88.75 |
| 05 | OpenVLAOSS | Stanford / Google DeepMind / TRI | Jun 2024 | OpenVLA: An Open-Source Vision-Language-Action Model · code | 76.50 |
Every paper below corresponds to at least one row in the leaderboard above. Click through for the arXiv preprint and, when available, the reference implementation.
Submit a checkpoint and a reproduction script. We will run it, publish the score, and — if it takes the top — annotate the step on the progress chart with your name.