VLA Harness MIKASA-Robo: leaderboard
Memory-focused robot manipulation where a policy must retain information about objects and goals that are no longer visible in the current observation.
Metric: Success Rate (%). Source: allenai.github.io. 9 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | MemoryVLA | 44.4 | |
| 2 | Octo-small | 31.6 | |
| 3 | pi-0 | 29.4 | |
| 4 | Diffusion Policy | 29 | |
| 5 | OpenVLA-OFT | 28.4 | |
| 6 | BC | 22 | |
| 7 | SpatialVLA | 21 | |
| 8 | CronusVLA | 18 | |
| 9 | OpenVLA | 11.2 |
Interactive version: theaggregate.ai/benchmark?slug=vla-harness-mikasa-robo · How It Works · Data refreshed daily, snapshot 2026-10-20.