Paper-reported
Ai2Maintainer ↗LIBERO's four suites (spatial, object, goal, long), SimplerEnv robots, CALVIN multi-step chains.
As of November 17, 2026, X-VLA leads with 98.1, followed by OpenVLA-OFT (97.1) and GR00T N1.7 (97.0). Official leaderboard ↗
Leaderboard
- Tasks
- 105
- Attempts / task
- —
- Median cost / task
- $0.0086
- Median time / task
- 9.5 s
Median of papers' reports, or the model's own paper. Only models matched on other boards are shown.
X-VLA98.198.298.697.8$0.0083est.9.1 sest.2
OpenVLA-OFT97.197.698.497.9$0.0084est.9.2 sest.3
GR00T N1.797.097.798.098.0$0.0084est.9.2 sest.4
π0.596.898.498.297.8$0.0084est.9.2 sest.5
UniVLA95.296.596.895.6$0.0085est.9.3 sest.6
GR00T N1.593.986.490.285.2$0.0086est.9.4 sest.7
π093.896.697.895.0$0.0086est.9.5 sest.8
RIPT-VLA93.692.795.698.4$0.0087est.9.5 sest.9
CogACT93.597.098.090.2$0.0087est.9.5 sest.10
SmolVLA87.290.096.092.0$0.0092est.10 sest.11
π0-FAST85.596.496.888.6$0.0093est.10 sest.12
OpenVLA76.584.788.479.2$0.010est.11 sest.13
Diffusion Policy72.478.388.968.3$0.010est.11 sest.Tasks
105 tasks. Hover a task to play its clip (105 of 105).
▶ Clip
▶ Clip
▶ Clip
▶ Clip
▶ Clip
▶ Clip
▶ Clip
▶ Clip
▶ Clip
▶ Clip
▶ Clip
▶ Clip
▶ Clip
▶ Clip
▶ Clip
▶ Clip
▶ Clip
▶ Clip
▶ Clip
▶ Clip
▶ Clip
▶ Clip
▶ Clip
▶ Clip
▶ Clip
▶ Clip
▶ Clip
▶ Clip
▶ Clip
▶ Clip
▶ Clip
▶ Clip
▶ Clip
▶ Clip
▶ Clip
▶ Clip
▶ Clip
▶ Clip
▶ Clip
▶ Clip
▶ Clip
▶ Clip
▶ Clip
▶ Clip
▶ Clip
▶ Clip
▶ Clip
▶ ClipMethodology
- Scoring
- Code check
- Time cap / attempt
- 18 s
- Control rate
- —
- Models in index
- 13 of 13
Scoring. Each paper's standard simulator checks; CALVIN counts completed subtasks.
Who runs it. Self-reported in papers, collected by AllenAI.
Robots. Simulated Franka, Google Robot, WidowX.
Note. Only models matched elsewhere are used. Lowest weight in the ranking.
Time per attempt. Not published. Estimated otherwise: time cap 18 s × (1 − success/2).
Cost per attempt. Not published. Estimated otherwise: one H100 at $3.29 per hour (Lambda on-demand, Oct 2026) × time. GPU price source ↗