VLA-Arena Long Horizon: leaderboard
Multi-stage manipulation where a policy must chain several sub-goals without losing track of progress, the hardest VLA-Arena category by a wide margin.
Metric: Success Rate (%). Source: vla-arena.github.io. 30 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | ACoT-VLA | 43.67 | |
| 2 | Paimon | 41.89 | |
| 3 | MolmoAct2 | 34.45 | |
| 4 | LingBot-VLA | 34 | |
| 5 | Qwen3-VL-Fast | 33.33 | |
| 6 | StellaVLA | 33.33 | |
| 7 | LangForce | 33 | |
| 8 | Qwen2.5-VL-GR00T | 32.92 | |
| 9 | Qwen3-VL-OFT | 32.67 | |
| 10 | LingBot-VLA 2.0 | 32.67 |
Interactive version: theaggregate.ai/benchmark?slug=vla-arena-long-horizon · How It Works · Data refreshed daily, snapshot 2026-10-20.