Models · Robot policy

Can it actually do the task on a real arm?

Models that output motor commands directly (VLA): given camera images and an instruction, they move the arm.

RoboArena

Rating from pairwise preferences (hover for ± std and number of evaluations); policies with fewer than 20 evaluations hidden

#1 · RoboArenaSpirit v1.61,788

1Spirit v1.61,788

2DreamZero1,735

3Apricot1,661

4π0.5-DROID1,608

5π0-FAST-DROID1,582

6G0.5-Droid-AR1,559

7PaliGemma VQ (DROID)1,546

8floppy_bagel_525k1,546

Source: robo-arena.github.io · snapshot 2026-10-28 · 7 universities, DROID platform · Results as published by each leaderboard; we have not rerun them

RoboChallenge Table30

Table30 V2 success rate, % (official ranking order); top score 56.80

#1 · RoboChallenge Table30LoopWAM43.33%

1LoopWAM43.33%

2VLA-DM0.5 (KDDI Research)40.67%

3ambase (AVerse001)31.67%

4my16 (Tymtbo)30.67%

5JV0 (JIIOV)16%

6mc (mcBrains)15.33%

7MagicBot14.67%

8Baseline π0.5 (RoboChallenge)14.33%

Source: robochallenge.cn · snapshot 2026-10-28 · Run by Dexmal + Hugging Face · Results as published by each leaderboard; we have not rerun them

SimFoundry

Real-world avg. success, %

#1 · SimFoundryπ0.573.1%

1π0.573.1%

2GR00T N1.757%

3DreamZero53%

4π042.3%

5GR00T N1.624.6%

Source: arXiv 2606.28276 · snapshot 2026-09-18 · Paper authors · Results as published by each leaderboard; we have not rerun them

FlagEval-Robo

Sim board: mean success of LIBERO, ManiSkill, VLA-Arena, RoboCasa, % (DreamZero unranked, two benchmarks only)

#1 · FlagEval-Roboπ0.566.88%

1π0.566.88%

2GR00T N1.663.21%

3LingBot-VLA-V263.13%

4ACoT-VLA61.22%

5π059.95%

6GR00T N1.759.34%

7Qwen-GR00T57.76%

8LingBot-VLA55.8%

9MolmoAct249.27%

10OpenVLA-OFT48.72%

11Motus48.68%

12SmolVLA29.92%

Source: flagevalrobo.baai.ac.cn · snapshot 2026-08 · Run by BAAI · Results as published by each leaderboard; we have not rerun them

ArmnetBench v0.1

Strict success, %

#1 · ArmnetBench v0.1π0.547.6%

1π0.547.6%

2π035.1%

3GR00T N1.729.4%

4Diffusion Policy26.7%

5ACT19.2%

6MolmoAct218.9%

7SmolVLA15%

Source: arXiv 2607.24481 · snapshot 2026-08-29 · Paper authors · Results as published by each leaderboard; we have not rerun them

RoboLab-120

Success rate, %

#1 · RoboLab-120Cosmos3 Nano Policy36.8%

1Cosmos3 Nano Policy36.8%

2Phoenix34.4%

3π0.528%

4DreamZero25.7%

5Cosmos3 Edge Policy22.9%

6π0-FAST15.5%

7GR00T N1.67.2%

8π05%

Source: RoboLab data · snapshot 2026-08-12 · Run by NVIDIA SRL · Results as published by each leaderboard; we have not rerun them

BEHAVIOR Challenge 2025

Held-out Q-score

#1 · BEHAVIOR Challenge 2025Robot Learning Collective0.26

1Robot Learning Collective0.26

2Comet (NVIDIA)0.251

3SimpleAI Robot0.159

4The North Star (Huawei)0.12

5Embodied Intelligence0.095

Source: BEHAVIOR leaderboard · snapshot 2025 final · Run by Stanford SVL · Results as published by each leaderboard; we have not rerun them

LIBERO

Four-suite avg. success, % (self-reported)

#1 · LIBEROSimpleVLA-RL (OFT)99.1%

1SimpleVLA-RL (OFT)99.1%

2Mixture of Horizons99%

3VLA-Adapter Pro98.5%

4ImageWAM98.4%

5RotVLA98.2%

6LaRA-VLA97.9%

Source: AllenAI vla-eval · snapshot 2026-09-23 · Self-reported in papers · Results as published by each leaderboard; we have not rerun them

Want your model on our bench, or a benchmark we don't cover yet? We run custom evaluations.Talk to us →