RoboCasa365 leaderboard

Household kitchen benchmark with 365 tasks across 2,500+ scenes. Leaderboard entries train only on the pretrain_human300 dataset; the 16 Composite-Unseen tasks are evaluated zero-shot and agentic entries are marked separately.

Simulation · Not stated on the leaderboard · MuJoCo (robosuite)Report an error
85.6%Best · HexaShell
29Models ranked
OpenSaturation
Nov 18, 2026Numbers checked
#ModelTypeScoreDateSource
1HexaShellHexaFutureAgent system85.6%Oct 31, 2026
2GR00T 2NVIDIAWorld-action67.2%, self-reported by the authors, not a leaderboard entryNov 2026
3Awomo-0.5IAwomo-WAM TeamWorld-action64.4%Oct 9, 2026
4minionxnovaxbotVLA62.6%Oct 19, 2026
5Awomo-0.5VAwomo-WAM TeamWorld-action62.5%Oct 9, 2026
6Axiom-04Axiom RoboticsVLA61.6%Oct 1, 2026
7Paimon-0PrimotionVLA58.1%Sep 11, 2026
8WorldDreamer v5World AgentsWorld-action57.5%Oct 10, 2026
9Xiaomi-Robotics-1Xiaomi RoboticsVLA57.4%Jul 8, 2026
10Phasor-m7PhasorVLA56.8%Sep 15, 2026
11Meituan-Robotics-0Meituan RoboticsVLA55.0%Oct 24, 2026
12LoopWAMLoopWAM TeamWorld-action48.5%Oct 20, 2026
13ProWAMMetaWorld-action48.1%Sep 30, 2026
14PatchWAMPatchWAM TeamWorld-action47.6%Sep 29, 2026
15ABot-M0.6AMAP CV LabVLA46.6%Jun 26, 2026
16ABot-M0.5AMAP CV LabWorld-action40.3%Jun 29, 2026
17RLDX-1.1RLWRLDVLA40.2%, self-reported by the authors, not a leaderboard entryNov 12, 2026
18PRTSTeleAIVLA39.6%Aug 18, 2026
19RLDX-1RLWRLDVLA36.0%May 20, 2026
20Qwen-RobotManipAlibaba QwenVLA35.9%, self-reported by the authors, not a leaderboard entryJun 2026
21WorldDreamerWorld AgentsWorld-action35.3%Jun 20, 2026
22Qwen-RobotManip-ContextAlibaba QwenVLA33.8%, self-reported by the authors, not a leaderboard entryJun 2026
23GR00T N1.5NVIDIAVLA23.9%May 19, 2026
24LY-GWM-RoboCasaSEIN-LYGWMVLA23.6%Sep 24, 2026
25GR00T N1.6NVIDIAVLA21.9%May 14, 2026
26GigaWorld-Policy 0.1GigaAIWorld-action20.7%May 11, 2026
27π0.5Physical IntelligenceVLA16.9%Apr 2, 2026
28π0Physical IntelligenceVLA14.8%Apr 2, 2026
29Diffusion PolicyColumbia / Toyota Research InstitutePolicy6.1%Apr 2, 2026

Metric: Mean task success over 50 target tasks in 3 splits (Atomic-Seen, Composite-Seen, Composite-Unseen). One result per model: RoboFYI re-run, then official leaderboard, then paper, then third party. Runs that deviate from the official protocol are listed on the model page but not ranked.

Wrong or missing number? Report an error