Third-party leaderboards

Ten groups run robot models through their own tests and publish the ranking. Pick a leaderboard to see it. A simulation ranking is not a real-robot ranking.

Runs every model itself, in simulation and on its own robot arms.

AI MMLab Club / HKU MMLab (non-profit) · Open RoboDojo ↗

Test

#	Model	Made by	Score	Success
1	SLA-PreviewHKUST(GZ) & OLA	HKUST(GZ) & OLA	42.3	33.9%
2	InternW0-ΔShanghai AI Lab	Shanghai AI Lab	38.9	31.7%
3	OpenWAM-αOpenWAM Team	OpenWAM Team	37.6	24.4%
4	MWA-Phoenix-PreviewAnyverse & OpenPhoenix	Anyverse & OpenPhoenix	35.3	26.1%
5	Pi-05RoboDojo Team	RoboDojo Team	22.9	12.8%
6	InternVLA-A1RoboDojo Team	RoboDojo Team	12	7.2%
7	GalaxeaVLA (G0)RoboDojo Team	RoboDojo Team	9	4.4%
8	Xiaomi-Robotics-0RoboDojo Team	RoboDojo Team	7.9	3.9%
9	X-VLARoboDojo Team	RoboDojo Team	7.6	3.3%
10	GROOT-N1.7RoboDojo Team	RoboDojo Team	5.9	1.7%
11	Pi-0RoboDojo Team	RoboDojo Team	5.8	1.7%
12	StarVLARoboDojo Team	RoboDojo Team	4.1	1.7%
13	Spirit v1.5RoboDojo Team	RoboDojo Team	1.6	0.6%
14	DM0RoboDojo Team	RoboDojo Team	0	0%

#	Model	Made by	Score	Success
1	Q0-PreviewQoo Tech	Qoo Tech	74.2	68.9%
2	GO-3 Nano PreviewAGIBOT	AGIBOT	61.9	56.4%
3	PyRUA-LeanPKU DAGroup	PKU DAGroup	55.3	50.4%
4	HexaShell-PreviewHexaFuture	HexaFuture	53.2	46.9%
5	WorldScape-R²SIManifold AI	Manifold AI	46.6	39.4%
6	EmbodiedEvoEinsia.AI	Einsia.AI	45.1	39.0%
7	WALL-OSS-1.0-liteX Square Robot	X Square Robot	42.6	35.0%
8	VPP2RobotEra	RobotEra	39.3	32.3%
9	PhysicalRSIHKU MMLab & KAI	HKU MMLab & KAI	36.3	31.4%
10	Awomo-0.5Awomo	Awomo	35.3	29.6%
11	Simate-betaSimate	Simate	34.0	28.0%
12	Rex-M1 PreviewFT Lab & VisIncept	FT Lab & VisIncept	33.8	27.4%
13	HWM-0 PreviewHorizon Robotics	Horizon Robotics	32.8	27.2%
14	VPP2-PreviewRobotEra	RobotEra	31.4	25.6%
15	InternW0-ΔShanghai AI Lab	Shanghai AI Lab	30.8	23.9%
16	Liber-0 PreviewLiberAI	LiberAI	30.7	25.5%
17	Liber-0 LiteLiberAI	LiberAI	29.2	24.2%
18	GPT-6-AstraRoboDojo Team	RoboDojo Team	29.0	22.5%
19	DM0.5Dexmal	Dexmal	24.9	19.3%
20	ME-Brain-1.0Li Auto Inc.	Li Auto Inc.	21.7	16.0%
21	GalaxeaVLA (G0.5)Galaxea AI	Galaxea AI	20.2	14.9%
22	Xiaomi-Robotics-1Xiaomi Robotics	Xiaomi Robotics	20.1	13.9%
23	MachEmbodied-U0Li Auto Inc.	Li Auto Inc.	17.8	11.5%
24	OpenWAM-αOpenWAM Team	OpenWAM Team	17.2	11.9%
25	EWAMEWAM Team	EWAM Team	15.2	9.4%
26	Meituan-Robotics-0Meituan Robotics	Meituan Robotics	14.9	9.5%
27	HWM-CogWAMHorizon Robotics	Horizon Robotics	14.3	10.3%
28	Hy-Embodied-0.5-VLATencent Robotics X	Tencent Robotics X	13.1	8.8%
29	KinRTIIGroup	IIGroup	13.0	8.8%
30	SimpleMemVLASimpleMemVLA Team	SimpleMemVLA Team	12.6	9.3%
31	Spatial ForcingOpenHelix Robotics	OpenHelix Robotics	12.4	8.0%
32	Pi-05RoboDojo Team	RoboDojo Team	11.4	6.9%
33	InternVLA-A1.5InternVLA Team	InternVLA Team	11.2	7.1%
34	StarVLA-PI_v3StarVLA Team	StarVLA Team	10.8	7.5%
35	VLActStarVLA Team	StarVLA Team	10.7	7.6%
36	X-VLARoboDojo Team	RoboDojo Team	10.1	6.5%
37	MolmoAct2MolmoAct2 Team	MolmoAct2 Team	9.0	5.0%
38	StarVLA-GR00Tbuilt on GR00T N1.7StarVLA Team	StarVLA Team	8.3	4.7%
39	StarVLA-OFTbuilt on OpenVLA-OFTStarVLA Team	StarVLA Team	8.2	4.8%
40	X-WAMX-WAM Team	X-WAM Team	7.7	3.8%
41	Xiaomi-Robotics-0RoboDojo Team	RoboDojo Team	6.9	4.2%
42	StarVLARoboDojo Team	RoboDojo Team	6.4	3.2%
43	GigaWorld-Policy-0RoboDojo Team	RoboDojo Team	6.2	3.3%
44	GalaxeaVLA (G0)RoboDojo Team	RoboDojo Team	5.8	3.0%
45	LingBot-VLARoboDojo Team	RoboDojo Team	5.5	3.0%
46	EventVLASJTU ScaleLab	SJTU ScaleLab	5.0	2.8%
47	AHA-WAMSJTU ScaleLab	SJTU ScaleLab	4.8	2.4%
48	ABot-M0RoboDojo Team	RoboDojo Team	3.7	1.7%
49	Fast-WAMRoboDojo Team	RoboDojo Team	3.5	2.0%
50	Pi-0RoboDojo Team	RoboDojo Team	3.5	1.5%
51	DeepSeek-FlashRoboDojo Team	RoboDojo Team	3.0	1.9%
52	GROOT-N1.7RoboDojo Team	RoboDojo Team	2.9	1.3%
53	InternVLA-A1RoboDojo Team	RoboDojo Team	2.5	1.1%
54	SmolVLA (Single Task)RoboDojo Team	RoboDojo Team	1.8	0.8%
55	LDA-1BRoboDojo Team	RoboDojo Team	1.7	0.5%
56	GPT-5.5RoboDojo Team	RoboDojo Team	1.1	0.9%
57	ACT (Single Task)RoboDojo Team	RoboDojo Team	1.1	0.5%
58	GO-1RoboDojo Team	RoboDojo Team	1.0	0.5%
59	H-RDTRoboDojo Team	RoboDojo Team	0.7	0.1%
60	RDTRoboDojo Team	RoboDojo Team	0.5	0.1%
61	DM0RoboDojo Team	RoboDojo Team	0.5	0.1%
62	Dexora-1BRoboDojo Team	RoboDojo Team	0.4	0.0%
63	A1RoboDojo Team	RoboDojo Team	0.3	0.0%
64	Spirit v1.5RoboDojo Team	RoboDojo Team	0.2	0.1%
65	TinyVLARoboDojo Team	RoboDojo Team	0.2	0.1%
66	OpenVLA-OFTRoboDojo Team	RoboDojo Team	0.2	0.0%

Score: RoboDojo score, 0–100: partial credit for each step done (its ranking); its success rate is in the last column. How it's run: Scores come from RoboDojo's own evaluation system, not from the teams: simulation over 3 seeds with hidden test layouts, real-robot runs on RoboDojo's own arms.

Fetched 5 Nov 2026 · updated by RoboDojo 3 Nov 2026

Runs submitted models on its own farm of real robots, 30 tasks.

Dexmal, with Hugging Face and partners · Open RoboChallenge ↗

Test

#	Model	Made by	Score	Success
1	Era0Robotera	Robotera	76.3	64.3%
2	DM0Dexmal	Dexmal	72.2	62%
3	GigaBrain-0.1lyf	lyf	68.3	51.7%
4	Spirit-v1.5Spirit AI	Spirit AI	67.2	51%
5	pi0.5rc_baseline	rc_baseline	61.8	42.7%
6	Lira_generalistAtlas	Atlas	59.8	45%
7	wall-oss-v0.1Pushi Zhang	Pushi Zhang	55.3	35.3%
8	Atlas_generalistAtlas	Atlas	55.1	39.7%
9	DM0_generalistDexmal	Dexmal	49.1	37.3%
10	pi0rc_baseline	rc_baseline	46.4	28.3%
11	VLOSicheng Xie	Sicheng Xie	42.3	25.7%
12	X-VLAChongyang Xu	Chongyang Xu	34.8	21.3%
13	GR00Tbuilt on GR00T N1.7Sicheng Xie	Sicheng Xie	31.6	14.3%
14	pi05_generalistbuilt on π0.5wyf	wyf	31.3	17.7%
15	ZR-0_generalistAtlas	Atlas	29.2	12.7%
16	RDT-1Bzsz	zsz	28.8	15%
17	GR00T-MULTISicheng Xie	Sicheng Xie	28.4	13.3%
18	VLA_test_0119Xiangyue Wu	Xiangyue Wu	22	5%
19	cogacthsk	hsk	21.8	11.7%
20	pi0_generalistbuilt on π0wyf	wyf	20.2	9%
21	VLA_test_0204Xiangyue Wu	Xiangyue Wu	17.3	9%
22	openvla-oftgkf	gkf	8.7	5%

#	Model	Made by	Score	Success
1	LoopWAMLoopWAM	LoopWAM	56.8	43.3%
2	VLA-DM0.5KDDI Research	KDDI Research	54.4	40.7%
3	ambaseAVerse001	AVerse001	43.7	31.7%
4	my16Tymtbo	Tymtbo	41.3	30.7%
5	Baseline ModelRoboChallenge	RoboChallenge	31.5	14.3%
6	holobrainHRL	HRL	27.9	14.3%
7	JV0JIIOV	JIIOV	27.6	16%
8	mcmcBrains	mcBrains	23.6	15.3%
9	test_01VLA	VLA	22.7	12.3%
10	W0WMA	WMA	21.9	12%
11	MagicBotMagicBot	MagicBot	20.8	14.7%
12	r1Hua Xue	Hua Xue	20.1	8.3%
13	hcp-eHCP-E	HCP-E	19.2	10.3%
14	debuggers_vladebuggers	debuggers	18.7	11%
15	pi05_loomsmartVLA	smartVLA	18.4	9.3%
16	EgoVLAYuhao Chen	Yuhao Chen	17.8	6.3%
17	Astraeus_V1.7Yuze Xuan	Yuze Xuan	16.4	9%
18	metricsMetrics	Metrics	16.1	5.7%
19	USTB-LinketicZhenyang Wei	Zhenyang Wei	14.9	7.3%
20	Astraeus_V1.5Yuze Xuan	Yuze Xuan	13.9	8.7%
21	ARX5-multitaskHua Xue	Hua Xue	13	6.7%
22	MultiVLAVertex	Vertex	12.9	6%
23	victvictory team	victory team	12.7	7.3%
24	RoboRushRoboRush	RoboRush	12.3	6%
25	Astraeus_V1.6Yuze Xuan	Yuze Xuan	11.6	6.3%
26	VLA-AgentKDDI Research	KDDI Research	10.3	5%
27	BrainCYunJinBrains	YunJinBrains	10.1	3%
28	AVPAny-01	Any-01	9.6	5%
29	xyzxyz	xyz	8.8	5.3%
30	TengenX0.1Yimin Chen	Yimin Chen	8.7	2%
31	Astraeus_V1Yuze Xuan	Yuze Xuan	8.4	3.7%
32	UR5-multitaskHua Xue	Hua Xue	8	6%
33	smartVLAsmartVLA	smartVLA	7.3	4%
34	NGRobNext Gen Robot	Next Gen Robot	6.3	3.3%
35	ALOHA-multitaskHua Xue	Hua Xue	5.9	2%
36	DOS-W1-multitaskHua Xue	Hua Xue	5.5	3%
37	KETI_pi0.5KETI	KETI	5.4	2%
38	TvTTvT	TvT	5.1	2%
39	EchoMind MasterEchoMind Master	EchoMind Master	4.2	0.3%
40	happy_farmYZF	YZF	4.1	1.7%
41	UR5_PI05Zhicheng Dai	Zhicheng Dai	3.9	2.3%
42	RobotPioneerRobotPioneer	RobotPioneer	3.2	0.7%
43	pi-aloha-baseAVerse001	AVerse001	3	0%
44	moveon能动就行	能动就行	2.6	0%
45	wpzhaogs	zhaogs	2.5	0%
46	pi05_basesmartVLA	smartVLA	2.3	1.3%
47	R0VLM	VLM	1.9	0%
48	Aloha05db0	db0	1.5	1%
49	WAM_v0Nanfang Zheng	Nanfang Zheng	1.4	0%
50	M36-E0Bo Zhang	Bo Zhang	1.2	0%
51	pi05robocare	robocare	0.7	0%
52	EI-WAM-53MJiahao Wu	Jiahao Wu	0.4	0%
53	Datawhale-VLADatawhale-EAI	Datawhale-EAI	0.2	0%
54	testmulti6THUEE	THUEE	0.2	0%
55	verticalVertical	Vertical	0.2	0%
56	pb_vlaPB	PB	0	0%
57	kp-robotLutos_sanzu	Lutos_sanzu	0	0%
58	DM-OPDWenjuan Han	Wenjuan Han	0	0%
59	MiniCPM-RobotManipModelbest	Modelbest	0	0%
60	objsightZhe Tao	Zhe Tao	0	0%

Score: RoboChallenge score, 0–100: partial credit for each step done (its ranking); its success rate is in the last column. How it's run: Teams submit a model; RoboChallenge runs it on its own robot farm and scores every run. Version 1 is no longer updated; version 2 is live.

Fetched 5 Nov 2026

Runs every model itself on a real arm, blind, with each result reviewed.

Positronic Robotics, with Nebius · Open PhAIL ↗

Test

#	Model	Made by	Success	Range
1	Cosmos3 NanoNVIDIA	NVIDIA	69.2%	64.1–74.3
2	FLUX 3 ActionBlack Forest Labs	Black Forest Labs	62.9%	57.8–68
3	MolmoAct2Ai2	Ai2	47%	42.1–51.9
4	G0.5Galaxea	Galaxea	39.1%	34.1–44.1
5	GR00Tbuilt on GR00T N1.7NVIDIA	NVIDIA	36%	31.2–40.8

#	Model	Made by	Success	Range
1	π₀.₅Physical Intelligence	Physical Intelligence	75.1%	70.3–79.9
2	GR00Tbuilt on GR00T N1.7NVIDIA	NVIDIA	72.3%	67.7–76.9
3	ACTAction Chunking Transformer	Action Chunking Transformer	63.7%	56.7–70.7
4	SmolVLAHugging Face	Hugging Face	32%	24.1–39.9

Score: Reliability, % of the task done within 240 s (± is half the 95% range). How it's run: Positronic runs every model itself on its own arm, with checkpoints rotated blind; each verdict is reviewed by an annotator. A person teleoperating the robot scores 100.

Fetched 5 Nov 2026 · updated by PhAIL 4 Nov 2026

People at 28 labs compare two models side by side on real robots and pick the better one.

UC Berkeley, Stanford and the DROID community · Open RoboArena ↗

#	Model	Made by	Score	Range
1	DreamZeroopen source	open source	1735	± 43
2	pi05_droidnot open source	not open source	1608	± 31
3	pi0_fast_droidbuilt on π0not open source	not open source	1582	± 30
4	paligemma_vq_droidopen source	open source	1546	± 30
5	paligemma_diffusion_droidopen source	open source	1533	± 30
6	paligemma_fast_specialist_droidopen source	open source	1531	± 30
7	paligemma_fast_droidopen source	open source	1516	± 30
8	pi0_droidnot open source	not open source	1461	± 30
9	paligemma_binning_droidopen source	open source	786	± 50

Score: Relative score from head-to-head comparisons (not a success rate); ± is the uncertainty RoboArena publishes. How it's run: Volunteers at 28 organisations run two policies side by side on their own DROID robots and pick the better one, blind. RoboArena turns the votes into a score. Only policies with 100 or more comparisons are shown.

Fetched 5 Nov 2026 · updated by RoboArena 5 Nov 2026

Retrains every model on the same data, then tests it in simulation and on real robots.

BAAI, Beijing Academy of Artificial Intelligence (non-profit) · Open FlagEval-Robo ↗

Test

#	Model	Made by	Score
1	Pi0.5Physical Intelligence	Physical Intelligence	56.7
2	Motus清华大学	清华大学	46
3	LingBot-VLA-v2蚂蚁灵波	蚂蚁灵波	43.9
4	LingBot-VLA蚂蚁灵波	蚂蚁灵波	33.8
5	GROOT N1.6built on GR00T N1.7NVIDIA	NVIDIA	32.6
6	SmolVLAHugging Face	Hugging Face	18.5

#	Model	Made by	Success	Runs
1	Pi0.5Physical Intelligence	Physical Intelligence	66.9%	7,350
2	GR00T N1.6built on GR00T N1.7NVIDIA	NVIDIA	63.2%	7,350
3	LingBot-VLA-V2蚂蚁灵波	蚂蚁灵波	63.1%	7,350
4	ACoT-VLA上海智元	上海智元	61.2%	7,350
5	Pi0Physical Intelligence	Physical Intelligence	59.9%	7,350
6	GR00T N1.7NVIDIA	NVIDIA	59.3%	7,350
7	Qwen-GR00Tbuilt on GR00T N1.7香港科技大学	香港科技大学	57.8%	7,300
8	LingBot-VLA蚂蚁灵波	蚂蚁灵波	55.8%	7,350
9	MolmoAct2Allen Institute for AI	Allen Institute for AI	49.3%	7,350
10	OpenVLA-OFT斯坦福大学	斯坦福大学	48.7%	7,350
11	Motus清华大学	清华大学	48.7%	7,300
12	SmolVLAHugging Face	Hugging Face	29.9%	7,350

#	Model	Made by	Success	Runs
1	LingBot-VLA-V2蚂蚁灵波	蚂蚁灵波	98.9%	2,000
2	Qwen-GR00Tbuilt on GR00T N1.7香港科技大学	香港科技大学	98.4%	2,000
3	LingBot-VLA蚂蚁灵波	蚂蚁灵波	98.3%	2,000
4	MolmoAct2Allen Institute for AI	Allen Institute for AI	98.1%	2,000
5	ACoT-VLA上海智元	上海智元	97.7%	2,000
6	Pi0.5Physical Intelligence	Physical Intelligence	97.6%	2,000
7	OpenVLA-OFT斯坦福大学	斯坦福大学	97.5%	2,000
8	GR00T N1.6built on GR00T N1.7NVIDIA	NVIDIA	96.4%	2,000
9	Pi0Physical Intelligence	Physical Intelligence	94.8%	2,000
10	GR00T N1.7NVIDIA	NVIDIA	94.0%	2,000
11	Motus清华大学	清华大学	81.7%	2,000
12	SmolVLAHugging Face	Hugging Face	60.8%	2,000

#	Model	Made by	Success	Runs
1	Pi0.5Physical Intelligence	Physical Intelligence	44.3%	400
2	ACoT-VLA上海智元	上海智元	38.3%	400
3	LingBot-VLA-V2蚂蚁灵波	蚂蚁灵波	37%	400
4	LingBot-VLA蚂蚁灵波	蚂蚁灵波	33.8%	400
5	GR00T N1.6built on GR00T N1.7NVIDIA	NVIDIA	33%	400
6	Pi0Physical Intelligence	Physical Intelligence	29.8%	400
7	GR00T N1.7NVIDIA	NVIDIA	29.8%	400
8	OpenVLA-OFT斯坦福大学	斯坦福大学	25.8%	400
9	Qwen-GR00Tbuilt on GR00T N1.7香港科技大学	香港科技大学	23.8%	400
10	Motus清华大学	清华大学	19.3%	400
11	SmolVLAHugging Face	Hugging Face	15.5%	400
12	MolmoAct2Allen Institute for AI	Allen Institute for AI	14.8%	400
13	DreamZeroNVIDIA	NVIDIA	6.5%	400

#	Model	Made by	Success	Runs
1	GR00T N1.6built on GR00T N1.7NVIDIA	NVIDIA	52.6%	1,100
2	Pi0.5Physical Intelligence	Physical Intelligence	46.7%	1,100
3	GR00T N1.7NVIDIA	NVIDIA	44.5%	1,100
4	Motus清华大学	清华大学	44.3%	1,100
5	Pi0Physical Intelligence	Physical Intelligence	42.8%	1,100
6	LingBot-VLA-V2蚂蚁灵波	蚂蚁灵波	41.6%	1,100
7	Qwen-GR00Tbuilt on GR00T N1.7香港科技大学	香港科技大学	36.6%	1,100
8	ACoT-VLA上海智元	上海智元	30.3%	1,100
9	LingBot-VLA蚂蚁灵波	蚂蚁灵波	15.8%	1,100
10	OpenVLA-OFT斯坦福大学	斯坦福大学	15.6%	1,100
11	SmolVLAHugging Face	Hugging Face	13.6%	1,100
12	MolmoAct2Allen Institute for AI	Allen Institute for AI	1.6%	1,100

#	Model	Made by	Success	Runs
1	MolmoAct2Allen Institute for AI	Allen Institute for AI	82.7%	3,850
2	Pi0.5Physical Intelligence	Physical Intelligence	79.0%	3,850
3	ACoT-VLA上海智元	上海智元	78.7%	3,850
4	LingBot-VLA蚂蚁灵波	蚂蚁灵波	75.4%	3,850
5	LingBot-VLA-V2蚂蚁灵波	蚂蚁灵波	75.0%	3,850
6	Pi0Physical Intelligence	Physical Intelligence	72.4%	3,850
7	Qwen-GR00Tbuilt on GR00T N1.7香港科技大学	香港科技大学	72.3%	3,800
8	GR00T N1.6built on GR00T N1.7NVIDIA	NVIDIA	70.8%	3,850
9	GR00T N1.7NVIDIA	NVIDIA	69.0%	3,850
10	OpenVLA-OFT斯坦福大学	斯坦福大学	56%	3,850
11	Motus清华大学	清华大学	49.5%	3,800
12	DreamZeroNVIDIA	NVIDIA	31.9%	3,850
13	SmolVLAHugging Face	Hugging Face	29.7%	3,850

Score: Real robot: share of task steps done, mean over both robots. Simulation: success rate, % of episodes; overall is the mean of the four benchmarks (BAAI's method, computed by us). How it's run: BAAI trains every model itself on the same data, from official code and weights, checks for test contamination, and runs all of it: simulation and real robot.

Fetched 27 Oct 2026

Runs every attempt itself on a real arm, mostly with AI chat models driving the robot.

Robocurve PBC · Open Robocurve ↗

#	Model	Made by	Success	Runs
1	Dream-1robot model	robot model	59.3%	120
2	Gemini 4 Argon (high)language model as robot agent	language model as robot agent	48.2%	120
3	GPT-6 Astra (high)language model as robot agent	language model as robot agent	42.5%	120
4	GPT-6.1 Terra (max)language model as robot agent	language model as robot agent	42.2%	120
5	GPT-6.1 Sol (high)language model as robot agent	language model as robot agent	41.7%	120
6	Opus 5.5 (high)language model as robot agent	language model as robot agent	38.2%	120
7	GPT-6 Astra (medium)language model as robot agent	language model as robot agent	36.7%	120
8	Gemini 3.8 Flash (high)language model as robot agent	language model as robot agent	36.2%	120
9	Opus 5.5 (medium)language model as robot agent	language model as robot agent	36%	120
10	GPT-6 Sol (high)language model as robot agent	language model as robot agent	22%	120
11	MolmoAct2robot model	robot model	18.9%	120
12	GPT-5.6 Sol (high)language model as robot agent	language model as robot agent	14.9%	120
13	GPT-5.6 Terra (max)language model as robot agent	language model as robot agent	11.4%	120
14	GR00T N1.7robot model	robot model	9.5%	120
15	Kimi K3.1 Pro (high)language model as robot agent	language model as robot agent	9.3%	120

Score: Mean task score, % (partial credit), 120 attempts per model. How it's run: Robocurve runs every attempt itself, mostly with large language models driving the robot; robot models are run as baselines. Every attempt has a log and three camera videos.

Fetched 5 Nov 2026 · updated by Robocurve 29 Sept 2026

Runs every model about 1,000 times per task in simulated homes.

Ai2 (Allen Institute for AI) · Open MolmoSpaces ↗

Test

#	Model	Made by	Success	Range
1	MagicBrain + Qwen3.7-PlusMagicLab	MagicLab	94.4%	92.8–95.7
2	MolmoBot-ImgAi2	Ai2	92.8%	91–94.2
3	MolmoBotAi2	Ai2	92.0%	90.2–93.5
4	MolmoBot-f3Ai2	Ai2	91.9%	90.1–93.5
5	Awomo-0.5Vawomo	awomo	86.4%	84.1–88.4
6	EchoPolicy-0 + Gemini-3.8-FlashMagicLab	MagicLab	84.7%	82.3–86.8
7	Awomo-0.5Iawomo	awomo	84.5%	82.1–86.6
8	PrimeR0Lumos	Lumos	77.5%	74.8–80
9	TGQWENUSTB	USTB	74.8%	72–77.4
10	WALL-OSS-0.5X Square Robot	X Square Robot	73.8%	71–76.4
11	Phoenix + Gemini-ERGeodesic	Geodesic	73.7%	70.9–76.3
12	Psi-R2PsiBot	PsiBot	73.1%	70.3–75.8
13	AsterPickTUM	TUM	72.3%	69.4–75
14	TiPToP + Gemini-ERMIT & UPenn	MIT & UPenn	68.7%	65.8–71.5
15	Cosmos3-Nano-PolicyNVIDIA	NVIDIA	65.5%	62.5–68.4
16	MolmoBot-π₀built on π0Ai2	Ai2	64%	61–66.9
17	Cosmic + Gemini-ERGeodesic	Geodesic	62.6%	59.6–65.6
18	Magic-PolicyMagic	Magic	60.5%	57.4–63.5
19	Vana-PolicyVana	Vana	59.7%	56.6–62.7
20	CAPpick + Gemini-ERNYU & Berkeley	NYU & Berkeley	55.4%	52.3–58.5
21	PRTS-DroidTeleAI	TeleAI	53.8%	50.7–56.9
22	DreamZero DROIDNVIDIA GEAR	NVIDIA GEAR	52.1%	49–55.1
23	MolmoAct2 DROIDAi2	Ai2	48.3%	45.2–51.4
24	π₀.₅-MolmoSpacesbuilt on π0.5Ai2	Ai2	45%	41.9–48.1
25	π₀.₅ DROIDPI	PI	36.4%	33.5–39.4
26	LAP-VLAPrinceton	Princeton	24.9%	22.3–27.7
27	π₀-FAST DROIDbuilt on π0PI	PI	22%	19.5–24.7
28	π₀ DROIDPI	PI	16.2%	14.1–18.6
29	StereoVLAGalbot / Peking University	Galbot / Peking University	7.0%	5.1–9.7

#	Model	Made by	Success	Range
1	MagicBrain + Qwen3.7-PlusMagicLab	MagicLab	84.7%	82.3–86.8
2	Awomo-0.5Iawomo	awomo	71%	68.1–73.7
3	Awomo-0.5Vawomo	awomo	58.9%	55.8–61.9
4	EchoPolicy-0 + Gemini-3.8-FlashMagicLab	MagicLab	55.6%	52.5–58.7
5	PrimeR0Lumos	Lumos	54.6%	51.5–57.7
6	WALL-OSS-0.5X Square Robot	X Square Robot	46%	42.9–49.1
7	Cosmic + Gemini-ERGeodesic	Geodesic	44.7%	41.6–47.8
8	Phoenix + Gemini-ERGeodesic	Geodesic	43.8%	40.8–46.9
9	Psi-R2PsiBot	PsiBot	41.5%	38.5–44.6
10	Vana-PolicyVana	Vana	36.1%	33.2–39.2
11	Cosmos3-Nano-PolicyNVIDIA	NVIDIA	35.2%	32.3–38.2
12	TiPToP + Gemini-ERMIT & UPenn	MIT & UPenn	33.2%	30.4–36.2
13	PRTS-DroidTeleAI	TeleAI	27.7%	25–30.6
14	DreamZero DROIDNVIDIA GEAR	NVIDIA GEAR	25.6%	22.9–28.4
15	MolmoAct2 DROIDAi2	Ai2	23.3%	20.8–26
16	π₀.₅ DROIDPI	PI	13.6%	11.6–15.9
17	π₀ DROIDPI	PI	12.5%	10.6–14.7
18	π₀-FAST DROIDbuilt on π0PI	PI	10.9%	9.2–13.1
19	LAP-VLAPrinceton	Princeton	6.6%	5.2–8.3

#	Model	Made by	Success	Range
1	EchoPolicy-0 + Gemini-3.8-FlashMagicLab	MagicLab	70.2%	67.3–73
2	Awomo-0.5Vawomo	awomo	62.4%	59.4–65.4
3	Cosmic + Gemini-ERGeodesic	Geodesic	58.5%	55.4–61.5
4	Awomo-0.5Iawomo	awomo	49.3%	46.2–52.4
5	CAPopen + Gemini-ERNYU & Berkeley	NYU & Berkeley	41.9%	38.9–45
6	Phoenix + Gemini-ERGeodesic	Geodesic	37.3%	34.4–40.3
7	ZoeInside-VLAZoellent AI	Zoellent AI	36.6%	33.7–39.6
8	Vana-PolicyVana	Vana	35.1%	32.2–38.1
9	Cosmos3-Nano-PolicyNVIDIA	NVIDIA	32%	29.2–35
10	PrimeR0Lumos	Lumos	27.8%	25.1–30.7
11	MagicBrain + Qwen3.7-PlusMagicLab	MagicLab	27.7%	25–30.6
12	DreamZero DROIDNVIDIA GEAR	NVIDIA GEAR	24.9%	22.3–27.6
13	π₀.₅ DROIDPI	PI	22.7%	20.2–25.4
14	WALL-OSS-0.5X Square Robot	X Square Robot	16.8%	14.6–19.2
15	PRTS-DroidTeleAI	TeleAI	12%	10.1–14.2
16	MolmoAct2 DROIDAi2	Ai2	11.7%	9.9–13.8
17	LAP-VLAPrinceton	Princeton	11.4%	9.9–13
18	π₀-FAST DROIDbuilt on π0PI	PI	11.1%	9.3–13.2
19	π₀ DROIDPI	PI	11%	9.2–13.1
20	Psi-R2PsiBot	PsiBot	9%	7.4–10.9

#	Model	Made by	Success	Range
1	EchoPolicy-0 + Gemini-3.8-FlashMagicLab	MagicLab	91.3%	89.3–92.9
2	Cosmic + Gemini-ERGeodesic	Geodesic	87.9%	85.6–89.8
3	Awomo-0.5Iawomo	awomo	87.2%	84.9–89.2
4	Awomo-0.5Vawomo	awomo	86.0%	83.6–88.1
5	ZoeInside-VLAZoellent AI	Zoellent AI	85.7%	83.3–87.8
6	CAPclose + Gemini-ERNYU & Berkeley	NYU & Berkeley	83.8%	81.3–86.1
7	MagicBrain + Qwen3.7-PlusMagicLab	MagicLab	82.7%	80.2–85
8	PrimeR0Lumos	Lumos	81.3%	78.7–83.7
9	Cosmos3-Nano-PolicyNVIDIA	NVIDIA	79.1%	76.4–81.6
10	WALL-OSS-0.5X Square Robot	X Square Robot	78.5%	75.7–81
11	PRTS-DroidTeleAI	TeleAI	76.3%	73.4–78.9
12	MolmoAct2 DROIDAi2	Ai2	71.3%	68.2–74.1
13	π₀.₅ DROIDPI	PI	65.1%	62–68.2
14	Psi-R2PsiBot	PsiBot	62.2%	59–65.3
15	Vana-PolicyVana	Vana	60.8%	57.6–63.9
16	DreamZero DROIDNVIDIA GEAR	NVIDIA GEAR	60.3%	57.1–63.5
17	π₀ DROIDPI	PI	53.1%	49.9–56.3
18	Phoenix + Gemini-ERGeodesic	Geodesic	50.0%	46.7–53.2
19	LAP-VLAPrinceton	Princeton	45.9%	42.9–48.8
20	π₀-FAST DROIDbuilt on π0PI	PI	38.6%	35.5–41.8

#	Model	Made by	Success	Range
1	MolmoBotAi2	Ai2	93.5%	91.8–94.9
2	MolmoBot-ImgAi2	Ai2	92.2%	90.4–93.7
3	MolmoBot-f3Ai2	Ai2	91.3%	89.4–92.9
4	WALL-OSS-0.5X Square Robot	X Square Robot	78.4%	75.7–80.8
5	TGQWENUSTB	USTB	75.4%	72.6–78
6	ZoeInside-VLAZoellent AI	Zoellent AI	74.8%	72–77.4
7	Phoenix + Gemini-ERGeodesic	Geodesic	72.6%	69.8–75.3
8	TiPToP + Gemini-ERMIT & UPenn	MIT & UPenn	67.5%	64.5–70.3
9	Cosmos3-Nano-PolicyNVIDIA	NVIDIA	66.5%	63.5–69.4
10	MolmoBot-π₀built on π0Ai2	Ai2	66.2%	63.2–69.1
11	Cosmic + Gemini-ERGeodesic	Geodesic	61.1%	58–64.1
12	CAPpick + Gemini-ERNYU & Berkeley	NYU & Berkeley	55.4%	52.3–58.5
13	Psi-R2PsiBot	PsiBot	50.2%	47.1–53.3
14	π₀.₅-MolmoBot-FTbuilt on π0.5Ai2	Ai2	48.0%	45–51.2
15	PRTS-DroidTeleAI	TeleAI	47.5%	44.4–50.6
16	MolmoAct2 DROIDAi2	Ai2	43.4%	40.4–46.5
17	LAP-VLAPrinceton	Princeton	19.4%	17–22
18	π₀.₅ DROIDPI	PI	18.1%	15.8–20.6
19	X-VLAAIR	AIR	3.3%	2.4–4.5

#	Model	Made by	Success	Range
1	MolmoBotAi2	Ai2	66.8%	63.8–69.7
2	MolmoBot-f3Ai2	Ai2	63.8%	60.8–66.7
3	MolmoBot-ImgAi2	Ai2	63.5%	60.5–66.4
4	Phoenix + Gemini-ERGeodesic	Geodesic	58.6%	55.5–61.6
5	ZoeInside-VLAZoellent AI	Zoellent AI	51.8%	48.7–54.9
6	Cosmic + Gemini-ERGeodesic	Geodesic	51.5%	48.4–54.5
7	TiPToP + Gemini-ERMIT & UPenn	MIT & UPenn	50%	46.9–53.1
8	WALL-OSS-0.5X Square Robot	X Square Robot	47.9%	44.8–51
9	Cosmos3-Nano-PolicyNVIDIA	NVIDIA	32.3%	29.5–35.3
10	π₀.₅-MolmoBot-FTbuilt on π0.5Ai2	Ai2	28.3%	25.6–31.1
11	Psi-R2PsiBot	PsiBot	23.4%	20.9–26.1
12	MolmoAct2 DROIDAi2	Ai2	20.5%	18.1–23.1
13	PRTS-DroidTeleAI	TeleAI	18.1%	15.8–20.6
14	π₀.₅ DROIDPI	PI	6.4%	5–8.1
15	LAP-VLAPrinceton	Princeton	2.4%	1.6–3.6
16	X-VLAAIR	AIR	0.5%	0.3–1.2

#	Model	Made by	Success	Range
1	MolmoBotAi2	Ai2	64%	61–66.9
2	MolmoBot-ImgAi2	Ai2	61.4%	58.3–64.4
3	MolmoBot-f3Ai2	Ai2	59%	55.9–62
4	Phoenix + Gemini-ERGeodesic	Geodesic	56.8%	53.7–59.8
5	ZoeInside-VLAZoellent AI	Zoellent AI	55.9%	52.8–59
6	Cosmic + Gemini-ERGeodesic	Geodesic	49.1%	46–52.2
7	TiPToP + Gemini-ERMIT & UPenn	MIT & UPenn	48.5%	45.4–51.6
8	WALL-OSS-0.5X Square Robot	X Square Robot	47.2%	44.1–50.3
9	Cosmos3-Nano-PolicyNVIDIA	NVIDIA	32.2%	29.4–35.2
10	Psi-R2PsiBot	PsiBot	25.8%	23.2–28.6
11	π₀.₅-MolmoBot-FTbuilt on π0.5Ai2	Ai2	25.8%	23.1–28.6
12	MolmoAct2 DROIDAi2	Ai2	21.9%	19.5–24.6
13	PRTS-DroidTeleAI	TeleAI	20.5%	18.1–23.1
14	π₀.₅ DROIDPI	PI	7.0%	5.6–8.8
15	LAP-VLAPrinceton	Princeton	3.1%	2.2–4.4
16	X-VLAAIR	AIR	0.7%	0.4–1.3

#	Model	Made by	Success	Range
1	MolmoBotAi2	Ai2	63.7%	60.7–66.6
2	MolmoBot-f3Ai2	Ai2	62.7%	59.7–65.6
3	MolmoBot-ImgAi2	Ai2	62.1%	59.1–65.1
4	Phoenix + Gemini-ERGeodesic	Geodesic	58.1%	55–61.1
5	Cosmic + Gemini-ERGeodesic	Geodesic	53%	49.9–56.1
6	TiPToP + Gemini-ERMIT & UPenn	MIT & UPenn	47.8%	44.7–50.9
7	ZoeInside-VLAZoellent AI	Zoellent AI	45.5%	42.4–48.6
8	WALL-OSS-0.5X Square Robot	X Square Robot	30.1%	27.3–33
9	π₀.₅-MolmoBot-FTbuilt on π0.5Ai2	Ai2	29.7%	27–32.7
10	Psi-R2PsiBot	PsiBot	27.2%	24.5–30
11	Cosmos3-Nano-PolicyNVIDIA	NVIDIA	25.7%	23.1–28.5
12	PRTS-DroidTeleAI	TeleAI	21.6%	19.2–24.3
13	MolmoAct2 DROIDAi2	Ai2	11.1%	9.3–13.2
14	π₀.₅ DROIDPI	PI	8.0%	6.3–10
15	LAP-VLAPrinceton	Princeton	2.7%	1.9–3.9
16	X-VLAAIR	AIR	0.8%	0.4–1.5

#	Model	Made by	Success	Range
1	MolmoBotAi2	Ai2	66.5%	63.5–69.3
2	MolmoBot-f3Ai2	Ai2	65.4%	62.4–68.3
3	MolmoBot-ImgAi2	Ai2	62.9%	59.9–65.9
4	MolmoBot-π₀built on π0Ai2	Ai2	46.6%	42.7–50.6
5	Phoenix + Gemini-ERGeodesic	Geodesic	44.4%	41.4–47.5
6	π₀.₅-MolmoBot-FTbuilt on π0.5Ai2	Ai2	43.5%	37.6–49.6
7	Cosmic + Gemini-ERGeodesic	Geodesic	40.7%	37.7–43.8
8	ZoeInside-VLAZoellent AI	Zoellent AI	37.8%	34.8–40.8
9	TiPToP + Gemini-ERMIT & UPenn	MIT & UPenn	29.4%	26.7–32.3
10	Cosmos3-Nano-PolicyNVIDIA	NVIDIA	26%	23.4–28.8
11	PRTS-DroidTeleAI	TeleAI	17.4%	15.2–19.9
12	MolmoAct2 DROIDAi2	Ai2	15.6%	13.5–18
13	WALL-OSS-0.5X Square Robot	X Square Robot	13.8%	11.8–16.1
14	π₀.₅ DROIDPI	PI	12.0%	9.5–15
15	Psi-R2PsiBot	PsiBot	2.6%	1.8–3.8
16	LAP-VLAPrinceton	Princeton	1.6%	0.9–2.9
17	X-VLAAIR	AIR	0.1%	0–0.3

#	Model	Made by	Success	Range
1	Cosmic + Gemini-ERGeodesic	Geodesic	43.3%	40.3–46.4
2	TiPToP + Gemini-ERMIT & UPenn	MIT & UPenn	38%	35–41.1
3	π₀.₅-MolmoBot-FTbuilt on π0.5Ai2	Ai2	28.7%	24.8–32.9
4	MolmoBot-f3Ai2	Ai2	28.4%	25.7–31.2
5	MolmoBotAi2	Ai2	26.4%	23.7–29.2
6	MolmoBot-π₀built on π0Ai2	Ai2	24.5%	20.7–28.8
7	Cosmos3-Nano-PolicyNVIDIA	NVIDIA	22.9%	20.4–25.6
8	ZoeInside-VLAZoellent AI	Zoellent AI	21.2%	18.8–23.8
9	MolmoBot-ImgAi2	Ai2	21%	18.6–23.6
10	PRTS-DroidTeleAI	TeleAI	19.0%	16.7–21.6
11	MolmoAct2 DROIDAi2	Ai2	18.5%	16.2–21
12	WALL-OSS-0.5X Square Robot	X Square Robot	15.9%	13.8–18.3
13	Phoenix + Gemini-ERGeodesic	Geodesic	10.8%	9–12.9
14	π₀.₅ DROIDPI	PI	10.3%	7.4–14.1
15	LAP-VLAPrinceton	Princeton	3.4%	1.9–6.2
16	Psi-R2PsiBot	PsiBot	2.9%	2–4.1
17	X-VLAAIR	AIR	1.9%	1.2–3

#	Model	Made by	Success	Range
1	MolmoBot-ImgAi2	Ai2	67.8%	64.8–70.6
2	MolmoBotAi2	Ai2	67.8%	64.8–70.6
3	MolmoBot-f3Ai2	Ai2	66.1%	63.1–69
4	π₀.₅-MolmoBot-FTbuilt on π0.5Ai2	Ai2	48.3%	45.2–51.4
5	MolmoBot-π₀built on π0Ai2	Ai2	46.2%	43.1–49.4
6	Phoenix + Gemini-ERGeodesic	Geodesic	43.9%	40.9–47
7	Cosmic + Gemini-ERGeodesic	Geodesic	41.7%	38.7–44.8
8	ZoeInside-VLAZoellent AI	Zoellent AI	36.2%	33.3–39.2
9	Cosmos3-Nano-PolicyNVIDIA	NVIDIA	34.3%	31.4–37.3
10	TiPToP + Gemini-ERMIT & UPenn	MIT & UPenn	31.5%	28.7–34.5
11	MolmoAct2 DROIDAi2	Ai2	19.3%	17–21.9
12	WALL-OSS-0.5X Square Robot	X Square Robot	18.3%	16–20.8
13	PRTS-DroidTeleAI	TeleAI	18.2%	15.9–20.7
14	π₀.₅ DROIDPI	PI	10.4%	8.6–12.5
15	LAP-VLAPrinceton	Princeton	1.5%	0.9–2.5
16	Psi-R2PsiBot	PsiBot	1.5%	0.9–2.5
17	X-VLAAIR	AIR	0.9%	0.5–1.8

#	Model	Made by	Success	Range
1	DreamZero DROIDNVIDIA GEAR	NVIDIA GEAR	18%	15.7–20.5
2	MolmoAct2 DROIDAi2	Ai2	17.2%	15–19.7
3	π₀.₅ DROIDPI	PI	14.1%	12.1–16.4
4	π₀-FAST DROIDbuilt on π0PI	PI	8.2%	6.7–10.1
5	π₀ DROIDPI	PI	2.8%	2–4

#	Model	Made by	Success	Range
1	MolmoAct2 DROIDAi2	Ai2	14%	12–16.3
2	DreamZero DROIDNVIDIA GEAR	NVIDIA GEAR	13.8%	11.8–16.1
3	π₀.₅ DROIDPI	PI	11.3%	9.6–13.3
4	π₀-FAST DROIDbuilt on π0PI	PI	6.9%	5.5–8.7
5	π₀ DROIDPI	PI	2.8%	2–4

#	Model	Made by	Success	Range
1	π₀ DROIDPI	PI	0%	0–0.7
2	π₀-FAST DROIDbuilt on π0PI	PI	0%	0–0.7
3	π₀.₅ DROIDPI	PI	0%	0–0.7
4	DreamZero DROIDNVIDIA GEAR	NVIDIA GEAR	0%	0–0.7
5	MolmoAct2 DROIDAi2	Ai2	0%	0–0.7

#	Model	Made by	Success	Range
1	π₀.₅ DROIDPI	PI	12.1%	10.2–14.3
2	MolmoAct2 DROIDAi2	Ai2	9.0%	7.3–10.9
3	DreamZero DROIDNVIDIA GEAR	NVIDIA GEAR	7%	5.6–8.8
4	π₀ DROIDPI	PI	3.3%	2.4–4.6
5	π₀-FAST DROIDbuilt on π0PI	PI	2.2%	1.5–3.3

#	Model	Made by	Success	Range
1	π₀.₅ DROIDPI	PI	14.2%	12.2–16.5
2	MolmoAct2 DROIDAi2	Ai2	10.1%	8.4–12.2
3	DreamZero DROIDNVIDIA GEAR	NVIDIA GEAR	7.1%	5.7–8.9
4	π₀-FAST DROIDbuilt on π0PI	PI	5.6%	4.3–7.2
5	π₀ DROIDPI	PI	4.3%	3.2–5.7

#	Model	Made by	Success	Range
1	DreamZero DROIDNVIDIA GEAR	NVIDIA GEAR	26.6%	22.9–30.6
2	MolmoAct2 DROIDAi2	Ai2	23.6%	20.1–27.5
3	π₀.₅ DROIDPI	PI	19.4%	16.2–23.1
4	π₀-FAST DROIDbuilt on π0PI	PI	13.4%	10.7–16.7
5	π₀ DROIDPI	PI	9.4%	7.2–12.3

#	Model	Made by	Success	Range
1	π₀-FAST DROIDbuilt on π0PI	PI	0.2%	0.1–0.7
2	π₀.₅ DROIDPI	PI	0.2%	0.1–0.7
3	MolmoAct2 DROIDAi2	Ai2	0.2%	0.1–0.7
4	π₀ DROIDPI	PI	0.1%	0–0.6
5	DreamZero DROIDNVIDIA GEAR	NVIDIA GEAR	0.1%	0–0.6

#	Model	Made by	Success	Range
1	π₀.₅ DROIDPI	PI	0.3%	0.1–0.9
2	DreamZero DROIDNVIDIA GEAR	NVIDIA GEAR	0.3%	0.1–0.9
3	π₀ DROIDPI	PI	0%	0–0.4
4	π₀-FAST DROIDbuilt on π0PI	PI	0%	0–0.4
5	MolmoAct2 DROIDAi2	Ai2	0%	0–0.4

Score: Success rate, % of episodes. How it's run: Ai2 runs every policy itself in its own simulated homes: about 1,000 episodes per task, with a 95% interval on each score.

Fetched 5 Nov 2026

Checks every submitted result in a simulated kitchen.

RoboCasa team (UT Austin, NVIDIA) · Open RoboCasa365 ↗

Test

#	Model	Made by	Success
1	Awomo-0.5IAwomo-WAM Team	Awomo-WAM Team	64.4%
2	minionxnovaxbot	novaxbot	62.6%
3	Awomo-0.5VAwomo-WAM Team	Awomo-WAM Team	62.5%
4	Axiom-04Axiom Robotics	4Axiom Robotics	61.6%
5	Paimon-0Primotion	Primotion	58.1%
6	Xiaomi-Robotics-1Xiaomi Robotics	Xiaomi Robotics	57.4%
7	Phasor-m7Phasor	Phasor	56.8%
8	Meituan-Robotics-0Meituan Robotics	Meituan Robotics	55%
9	LoopWAMLoopWAM Team	LoopWAM Team	48.5%
10	ProWAMMeta	Meta	48.1%
11	PatchWAMPatchWAM Team	PatchWAM Team	47.6%
12	ABot-M0.6AMAP CV Lab	AMAP CV Lab	46.6%
13	ABot-M0.5AMAP CV Lab	AMAP CV Lab	40.3%
14	PRTSTeleAI	TeleAI	39.6%
15	RLDX-1RLWRLD	RLWRLD	36%
16	WorldDreamerWorld Agents	World Agents	35.3%
17	GR00T N1.5built on GR00T N1.7RoboCasa Team	RoboCasa Team	23.9%
18	LY-GWM-RoboCasaSEIN-LYGWM	SEIN-LYGWM	23.6%
19	GR00T N1.6built on GR00T N1.7RoboCasa Team	RoboCasa Team	21.9%
20	GigaWorld-Policy 0.1GigaAI	GigaAI	20.7%
21	π0.5RoboCasa Team	RoboCasa Team	16.9%
22	π0RoboCasa Team	RoboCasa Team	14.8%
23	Azero-Robotics-1GitHub: yuxin101	GitHub: yuxin101	12.6%
24	Diffusion PolicyRoboCasa Team	RoboCasa Team	6.1%

#	Model	Made by	Success
1	Awomo-0.5IAwomo-WAM Team	Awomo-WAM Team	86.5%
2	Axiom-04Axiom Robotics	4Axiom Robotics	86.5%
3	Awomo-0.5VAwomo-WAM Team	Awomo-WAM Team	85%
4	minionxnovaxbot	novaxbot	81.8%
5	Xiaomi-Robotics-1Xiaomi Robotics	Xiaomi Robotics	80.2%
6	ABot-M0.6AMAP CV Lab	AMAP CV Lab	79.4%
7	Paimon-0Primotion	Primotion	79.3%
8	PatchWAMPatchWAM Team	PatchWAM Team	78.7%
9	Phasor-m7Phasor	Phasor	78.7%
10	LoopWAMLoopWAM Team	LoopWAM Team	78%
11	Meituan-Robotics-0Meituan Robotics	Meituan Robotics	77.3%
12	ProWAMMeta	Meta	76%
13	ABot-M0.5AMAP CV Lab	AMAP CV Lab	75.6%
14	RLDX-1RLWRLD	RLWRLD	67.6%
15	PRTSTeleAI	TeleAI	66.3%
16	WorldDreamerWorld Agents	World Agents	66.3%
17	GR00T N1.6built on GR00T N1.7RoboCasa Team	RoboCasa Team	51.1%
18	GR00T N1.5built on GR00T N1.7RoboCasa Team	RoboCasa Team	50.7%
19	LY-GWM-RoboCasaSEIN-LYGWM	SEIN-LYGWM	48.3%
20	GigaWorld-Policy 0.1GigaAI	GigaAI	44.4%
21	π0.5RoboCasa Team	RoboCasa Team	39.6%
22	π0RoboCasa Team	RoboCasa Team	34.6%
23	Azero-Robotics-1GitHub: yuxin101	GitHub: yuxin101	30.3%
24	Diffusion PolicyRoboCasa Team	RoboCasa Team	15.7%

#	Model	Made by	Success
1	Awomo-0.5VAwomo-WAM Team	Awomo-WAM Team	60.4%
2	minionxnovaxbot	novaxbot	60%
3	Axiom-04Axiom Robotics	4Axiom Robotics	59.4%
4	Awomo-0.5IAwomo-WAM Team	Awomo-WAM Team	58.8%
5	Meituan-Robotics-0Meituan Robotics	Meituan Robotics	57.8%
6	Phasor-m7Phasor	Phasor	57.4%
7	Xiaomi-Robotics-1Xiaomi Robotics	Xiaomi Robotics	57.1%
8	Paimon-0Primotion	Primotion	55.8%
9	LoopWAMLoopWAM Team	LoopWAM Team	48.3%
10	ABot-M0.6AMAP CV Lab	AMAP CV Lab	48.3%
11	ProWAMMeta	Meta	46.5%
12	PatchWAMPatchWAM Team	PatchWAM Team	46.4%
13	ABot-M0.5AMAP CV Lab	AMAP CV Lab	37.7%
14	PRTSTeleAI	TeleAI	30.3%
15	RLDX-1RLWRLD	RLWRLD	27.9%
16	WorldDreamerWorld Agents	World Agents	26.7%
17	LY-GWM-RoboCasaSEIN-LYGWM	SEIN-LYGWM	16.3%
18	GR00T N1.5built on GR00T N1.7RoboCasa Team	RoboCasa Team	14.8%
19	GigaWorld-Policy 0.1GigaAI	GigaAI	11.8%
20	GR00T N1.6built on GR00T N1.7RoboCasa Team	RoboCasa Team	9.4%
21	π0.5RoboCasa Team	RoboCasa Team	7.1%
22	π0RoboCasa Team	RoboCasa Team	6.1%
23	Azero-Robotics-1GitHub: yuxin101	GitHub: yuxin101	3.8%
24	Diffusion PolicyRoboCasa Team	RoboCasa Team	0.2%

#	Model	Made by	Success
1	Awomo-0.5IAwomo-WAM Team	Awomo-WAM Team	45.2%
2	minionxnovaxbot	novaxbot	43.6%
3	Awomo-0.5VAwomo-WAM Team	Awomo-WAM Team	39.4%
4	Paimon-0Primotion	Primotion	36.7%
5	Axiom-04Axiom Robotics	4Axiom Robotics	35.8%
6	Xiaomi-Robotics-1Xiaomi Robotics	Xiaomi Robotics	32.1%
7	Phasor-m7Phasor	Phasor	31.5%
8	Meituan-Robotics-0Meituan Robotics	Meituan Robotics	27.1%
9	PRTSTeleAI	TeleAI	18.8%
10	ProWAMMeta	Meta	18.2%
11	LoopWAMLoopWAM Team	LoopWAM Team	15.4%
12	PatchWAMPatchWAM Team	PatchWAM Team	13.8%
13	WorldDreamerWorld Agents	World Agents	9%
14	RLDX-1RLWRLD	RLWRLD	8.5%
15	ABot-M0.6AMAP CV Lab	AMAP CV Lab	7.9%
16	ABot-M0.5AMAP CV Lab	AMAP CV Lab	3.3%
17	LY-GWM-RoboCasaSEIN-LYGWM	SEIN-LYGWM	3.1%
18	GigaWorld-Policy 0.1GigaAI	GigaAI	2.9%
19	GR00T N1.5built on GR00T N1.7RoboCasa Team	RoboCasa Team	2.7%
20	GR00T N1.6built on GR00T N1.7RoboCasa Team	RoboCasa Team	1.7%
21	Azero-Robotics-1GitHub: yuxin101	GitHub: yuxin101	1.6%
22	Diffusion PolicyRoboCasa Team	RoboCasa Team	1.3%
23	π0.5RoboCasa Team	RoboCasa Team	1.2%
24	π0RoboCasa Team	RoboCasa Team	1.1%

Score: Success rate, % of trials. How it's run: Teams send their results with code and checkpoint; the RoboCasa team re-runs or checks each one before listing it (closed models give it private access).

Fetched 5 Nov 2026

Runs models on 50 simulated two-arm tasks (only its own runs shown).

RoboTwin team (HKU MMLab, Tsinghua) · Open RoboTwin 2.0 ↗

Test

#	Model	Made by	Success	Runs
1	π0.5run by RoboTwin	run by RoboTwin	58.3%	5,000
2	Spatial Forcingrun by RoboTwin	run by RoboTwin	52%	5,000
3	X-WAMrun by RoboTwin	run by RoboTwin	47.9%	5,000
4	X-VLArun by RoboTwin	run by RoboTwin	44.5%	5,000
5	Abot-M0run by RoboTwin	run by RoboTwin	43.9%	5,000
6	EventVLArun by RoboTwin	run by RoboTwin	40.7%	5,000
7	Xiaomi Robotics-0run by RoboTwin	run by RoboTwin	40.5%	5,000
8	FastWAMrun by RoboTwin	run by RoboTwin	39.8%	5,000
9	GalaxeaVLArun by RoboTwin	run by RoboTwin	37.7%	5,000
10	AHA-WAMrun by RoboTwin	run by RoboTwin	33.8%	5,000
11	starVLArun by RoboTwin	run by RoboTwin	24.8%	5,000

#	Model	Made by	Success	Runs
1	π0.5run by RoboTwin	run by RoboTwin	46%	5,000
2	Abot-M0run by RoboTwin	run by RoboTwin	30.4%	5,000
3	Spatial Forcingrun by RoboTwin	run by RoboTwin	26.7%	5,000
4	X-WAMrun by RoboTwin	run by RoboTwin	25.8%	5,000
5	X-VLArun by RoboTwin	run by RoboTwin	20.9%	5,000
6	Xiaomi Robotics-0run by RoboTwin	run by RoboTwin	18.2%	5,000
7	EventVLArun by RoboTwin	run by RoboTwin	15.7%	5,000
8	GalaxeaVLArun by RoboTwin	run by RoboTwin	12.7%	5,000
9	AHA-WAMrun by RoboTwin	run by RoboTwin	3.2%	5,000
10	starVLArun by RoboTwin	run by RoboTwin	3.2%	5,000
11	FastWAMrun by RoboTwin	run by RoboTwin	1.9%	5,000

#	Model	Made by	Success	Runs
1	π0run by RoboTwin	run by RoboTwin	31.4%	5,000
2	DP3run by RoboTwin	run by RoboTwin	30.1%	5,000
3	RDTrun by RoboTwin	run by RoboTwin	24.1%	5,000
4	ACTrun by RoboTwin	run by RoboTwin	15.7%	5,000
5	DPrun by RoboTwin	run by RoboTwin	14.3%	5,000

#	Model	Made by	Success	Runs
1	π0run by RoboTwin	run by RoboTwin	16.3%	5,000
2	RDTrun by RoboTwin	run by RoboTwin	13.7%	5,000
3	DP3run by RoboTwin	run by RoboTwin	5.0%	5,000
4	ACTrun by RoboTwin	run by RoboTwin	1.7%	5,000
5	DPrun by RoboTwin	run by RoboTwin	0.6%	5,000

Score: Success rate, % of trials. How it's run: Only the rows the RoboTwin team ran itself are shown. Clean = the scene it was trained on; randomized = new clutter, lighting and textures.

Fetched 5 Nov 2026 · updated by RoboTwin 2.0 30 Oct 2026

Runs submitted models in its own simulator. Its own model is on the board.

AgiBot · Open RoboColiseum ↗

Test

#	Model	Made by	Score
1	PyRUA-Leanbaseline run by AgiBot	baseline run by AgiBot	82.8
2	GO-3 Nano PreviewAGIBOT	AGIBOT	78.8
3	Prime-0Primotion	Primotion	77.5
4	HiDream-O1-EmbodiedHiDream.ai	HiDream.ai	76.9
5	Claude Opus 5.5baseline run by AgiBot	baseline run by AgiBot	76.3
6	GPT-6 Astrabaseline run by AgiBot	baseline run by AgiBot	76
7	GPT-6.1 Solbaseline run by AgiBot	baseline run by AgiBot	70.9
8	DM0.5Dexmal	Dexmal	70.8
9	Gemini 4 Argonbaseline run by AgiBot	baseline run by AgiBot	66.9
10	GigaBrain-0.7极佳科技	极佳科技	65
11	WNDFCQingYuanChun	QingYuanChun	59.8
12	Xiaomi-Robotics-1智元机器人	智元机器人	59.4
13	OpenWAM-AlphaOpenWAM	OpenWAM	58.4
14	pi0.5_refinebuilt on π0.5宇宙小队	宇宙小队	57.9
15	π₀.₅baseline run by AgiBot	baseline run by AgiBot	57.4
16	kp-robot开普云信息科技股份有限公司	开普云信息科技股份有限公司	56.7
17	ACoT-VLAbaseline run by AgiBot	baseline run by AgiBot	56.3
18	lingbot-vla-v2baseline run by AgiBot	baseline run by AgiBot	55.8
19	DeepSeek-V4.1-Flashbaseline run by AgiBot	baseline run by AgiBot	55.1
20	Dream-1 SFT 500 stepsDreamLabs	DreamLabs	47.2
21	GR00T N1.7baseline run by AgiBot	baseline run by AgiBot	46.8
22	InternVLA-A1.5baseline run by AgiBot	baseline run by AgiBot	46.7
23	Xiaomi-Robotics-0baseline run by AgiBot	baseline run by AgiBot	43.1
24	Muse Wonder 1.0baseline run by AgiBot	baseline run by AgiBot	42.3
25	Dream-1 zero-shot智元	智元	33.5
26	π₀baseline run by AgiBot	baseline run by AgiBot	29

#	Model	Made by	Score
1	GPT-6 Astrabaseline run by AgiBot	baseline run by AgiBot	99.4
2	PyRUA-Leanbaseline run by AgiBot	baseline run by AgiBot	99.3
3	Claude Opus 5.5baseline run by AgiBot	baseline run by AgiBot	96.6
4	GPT-6.1 Solbaseline run by AgiBot	baseline run by AgiBot	95.5
5	Prime-0Primotion	Primotion	89.6
6	GO-3 Nano PreviewAGIBOT	AGIBOT	89.1
7	HiDream-O1-EmbodiedHiDream.ai	HiDream.ai	89.1
8	Gemini 4 Argonbaseline run by AgiBot	baseline run by AgiBot	85.9
9	DM0.5Dexmal	Dexmal	84.4
10	GigaBrain-0.7极佳科技	极佳科技	81.7
11	Xiaomi-Robotics-1智元机器人	智元机器人	77
12	ACoT-VLAbaseline run by AgiBot	baseline run by AgiBot	75.7
13	DeepSeek-V4.1-Flashbaseline run by AgiBot	baseline run by AgiBot	75.5
14	pi0.5_refinebuilt on π0.5宇宙小队	宇宙小队	75.4
15	π₀.₅baseline run by AgiBot	baseline run by AgiBot	74.6
16	kp-robot开普云信息科技股份有限公司	开普云信息科技股份有限公司	73.4
17	WNDFCQingYuanChun	QingYuanChun	71.4
18	lingbot-vla-v2baseline run by AgiBot	baseline run by AgiBot	70.8
19	OpenWAM-AlphaOpenWAM	OpenWAM	68.4
20	Dream-1 SFT 500 stepsDreamLabs	DreamLabs	67
21	InternVLA-A1.5baseline run by AgiBot	baseline run by AgiBot	65
22	GR00T N1.7baseline run by AgiBot	baseline run by AgiBot	64.6
23	Xiaomi-Robotics-0baseline run by AgiBot	baseline run by AgiBot	64.6
24	Muse Wonder 1.0baseline run by AgiBot	baseline run by AgiBot	58.6
25	Dream-1 zero-shot智元	智元	55.3
26	π₀baseline run by AgiBot	baseline run by AgiBot	36.8

#	Model	Made by	Score
1	PyRUA-Leanbaseline run by AgiBot	baseline run by AgiBot	97.9
2	Claude Opus 5.5baseline run by AgiBot	baseline run by AgiBot	82.5
3	GPT-6 Astrabaseline run by AgiBot	baseline run by AgiBot	78.3
4	GPT-6.1 Solbaseline run by AgiBot	baseline run by AgiBot	77.7
5	GO-3 Nano PreviewAGIBOT	AGIBOT	74.6
6	HiDream-O1-EmbodiedHiDream.ai	HiDream.ai	74.4
7	Gemini 4 Argonbaseline run by AgiBot	baseline run by AgiBot	69.4
8	Prime-0Primotion	Primotion	67.9
9	DM0.5Dexmal	Dexmal	61.5
10	OpenWAM-AlphaOpenWAM	OpenWAM	50.4
11	GigaBrain-0.7极佳科技	极佳科技	47.3
12	WNDFCQingYuanChun	QingYuanChun	43.7
13	DeepSeek-V4.1-Flashbaseline run by AgiBot	baseline run by AgiBot	43.5
14	lingbot-vla-v2baseline run by AgiBot	baseline run by AgiBot	42
15	pi0.5_refinebuilt on π0.5宇宙小队	宇宙小队	39.8
16	ACoT-VLAbaseline run by AgiBot	baseline run by AgiBot	39.7
17	π₀.₅baseline run by AgiBot	baseline run by AgiBot	35.6
18	Xiaomi-Robotics-1智元机器人	智元机器人	34.6
19	Muse Wonder 1.0baseline run by AgiBot	baseline run by AgiBot	31.9
20	kp-robot开普云信息科技股份有限公司	开普云信息科技股份有限公司	31.7
21	Dream-1 SFT 500 stepsDreamLabs	DreamLabs	29.8
22	InternVLA-A1.5baseline run by AgiBot	baseline run by AgiBot	27.4
23	GR00T N1.7baseline run by AgiBot	baseline run by AgiBot	24.9
24	Xiaomi-Robotics-0baseline run by AgiBot	baseline run by AgiBot	22.3
25	Dream-1 zero-shot智元	智元	16
26	π₀baseline run by AgiBot	baseline run by AgiBot	13

#	Model	Made by	Score
1	GO-3 Nano PreviewAGIBOT	AGIBOT	68.8
2	Prime-0Primotion	Primotion	66.1
3	HiDream-O1-EmbodiedHiDream.ai	HiDream.ai	64.4
4	DM0.5Dexmal	Dexmal	63.7
5	GigaBrain-0.7极佳科技	极佳科技	63.2
6	kp-robot开普云信息科技股份有限公司	开普云信息科技股份有限公司	60.2
7	Xiaomi-Robotics-1智元机器人	智元机器人	60.1
8	WNDFCQingYuanChun	QingYuanChun	59.6
9	π₀.₅baseline run by AgiBot	baseline run by AgiBot	58.2
10	pi0.5_refinebuilt on π0.5宇宙小队	宇宙小队	56.9
11	OpenWAM-AlphaOpenWAM	OpenWAM	56
12	PyRUA-Leanbaseline run by AgiBot	baseline run by AgiBot	49
13	ACoT-VLAbaseline run by AgiBot	baseline run by AgiBot	47.7
14	lingbot-vla-v2baseline run by AgiBot	baseline run by AgiBot	47
15	Claude Opus 5.5baseline run by AgiBot	baseline run by AgiBot	46.7
16	GPT-6 Astrabaseline run by AgiBot	baseline run by AgiBot	45.9
17	GR00T N1.7baseline run by AgiBot	baseline run by AgiBot	43.8
18	GPT-6.1 Solbaseline run by AgiBot	baseline run by AgiBot	40
19	Gemini 4 Argonbaseline run by AgiBot	baseline run by AgiBot	38.8
20	InternVLA-A1.5baseline run by AgiBot	baseline run by AgiBot	37
21	π₀baseline run by AgiBot	baseline run by AgiBot	34.7
22	DeepSeek-V4.1-Flashbaseline run by AgiBot	baseline run by AgiBot	33.3
23	Dream-1 SFT 500 stepsDreamLabs	DreamLabs	32.7
24	Xiaomi-Robotics-0baseline run by AgiBot	baseline run by AgiBot	30.7
25	Muse Wonder 1.0baseline run by AgiBot	baseline run by AgiBot	27.3
26	Dream-1 zero-shot智元	智元	20.7

#	Model	Made by	Score
1	Prime-0Primotion	Primotion	86.5
2	PyRUA-Leanbaseline run by AgiBot	baseline run by AgiBot	84.9
3	GO-3 Nano PreviewAGIBOT	AGIBOT	82.8
4	GPT-6 Astrabaseline run by AgiBot	baseline run by AgiBot	80.4
5	HiDream-O1-EmbodiedHiDream.ai	HiDream.ai	79.8
6	Claude Opus 5.5baseline run by AgiBot	baseline run by AgiBot	79.3
7	DM0.5Dexmal	Dexmal	73.4
8	Gemini 4 Argonbaseline run by AgiBot	baseline run by AgiBot	73.4
9	GPT-6.1 Solbaseline run by AgiBot	baseline run by AgiBot	70.4
10	DeepSeek-V4.1-Flashbaseline run by AgiBot	baseline run by AgiBot	68.1
11	GigaBrain-0.7极佳科技	极佳科技	68
12	Xiaomi-Robotics-1智元机器人	智元机器人	65.8
13	WNDFCQingYuanChun	QingYuanChun	64.6
14	lingbot-vla-v2baseline run by AgiBot	baseline run by AgiBot	63.3
15	ACoT-VLAbaseline run by AgiBot	baseline run by AgiBot	62.2
16	kp-robot开普云信息科技股份有限公司	开普云信息科技股份有限公司	61.4
17	π₀.₅baseline run by AgiBot	baseline run by AgiBot	61.3
18	pi0.5_refinebuilt on π0.5宇宙小队	宇宙小队	59.3
19	Dream-1 SFT 500 stepsDreamLabs	DreamLabs	59.2
20	OpenWAM-AlphaOpenWAM	OpenWAM	58.8
21	InternVLA-A1.5baseline run by AgiBot	baseline run by AgiBot	57.5
22	Xiaomi-Robotics-0baseline run by AgiBot	baseline run by AgiBot	54.6
23	GR00T N1.7baseline run by AgiBot	baseline run by AgiBot	53.8
24	Muse Wonder 1.0baseline run by AgiBot	baseline run by AgiBot	51.6
25	Dream-1 zero-shot智元	智元	42
26	π₀baseline run by AgiBot	baseline run by AgiBot	31.3

Score: AgiBot score, 0–100 (published as 0–1). How it's run: Teams expose their model; AgiBot runs the evaluation on its own servers and scores it. AgiBot also runs the well-known models as baselines, and its own model is on the board.

Fetched 5 Nov 2026 · updated by RoboColiseum 3 Nov 2026

Our scorecard models are in orange; lighter orange is built on one.

What the numbers say

Too few trials

On 18 of 22 tests, #1 and #2 are too close to call

Even with about 1,000 runs each. With 20 runs, a 70% model and an 80% model look the same:

20 trials

300 trials

70% model 80% model · each bar is where the true rate could be

Far from production

The best real-robot result is 75%. Production needs 95%+

Each model's best success rate on a real robot.

what production needs

No standard

π0.5 ranks #1 on one leaderboard and #21 of 24 on another

Each dot is one leaderboard. Every one uses its own robot, tasks and score.

topbottom

Simulation is not reality

Up to 93 points lost going from simulation to a real robot

A popular simulation test (LIBERO) against the same model's best real-robot result.

π0.5−23
SmolVLA−29
π0−67
OpenVLA-OFT−93
GR00T N1.7−92

simulation real robot

How we use these

Each leaderboard is one test, chosen by the group that runs it. A high rank means a model does well on that test, not that it is the best model for your work. We copy each group's published numbers, rank them the way they do, and link back to them.

AI Model Performance → · Methodology

Third-party leaderboards

What the numbers say

On 18 of 22 tests, #1 and #2 are too close to call

The best real-robot result is 75%. Production needs 95%+

π0.5 ranks #1 on one leaderboard and #21 of 24 on another

Up to 93 points lost going from simulation to a real robot

How we use these

You're subscribed

Get notified