Benchmark Leaderboards

Track which models hold the current SOTA on the benchmarks the field actually uses. Click a column header to sort; the leading row is highlighted.

4 tasks tracked

calvin

CALVIN long-horizon language-conditioned manipulation — average number of consecutive tasks completed (0–5) in the 1000-chain evaluation. Higher means longer successful task chains.

Metric: Avg Sequence Length — higher is better · Source ↗

# ▲	Model / Paper	Avg Sequence Length	Δ vs SOTA	Links
1	Xiaomi-Robotics-0SOTA	4.75	—	arxiv.org →
2	VITA	4.73	-0.02	arxiv.org →
3	NS-VLA	4.72	-0.03	arxiv.org →
4	AVA-VLA	4.65	-0.10	arxiv.org →
5	UD-VLA	4.64	-0.11	arxiv.org →
6	OASIS	4.57	-0.18	arxiv.org →
7	FLOWER + PTR	4.56	-0.19	arxiv.org →
8	MoLA	4.55	-0.20	arxiv.org →
9	ThinkProprio	4.55	-0.20	arxiv.org →
10	FLOWER (from PTR)	4.54	-0.21	arxiv.org →
11	UD-VLA + Fast-dVLA	4.54	-0.21	arxiv.org →
12	FLOWER	4.53	-0.22	arxiv.org →
13	MDT (from UD-VLA + Fast-dVLA)	4.52	-0.23	arxiv.org →
14	DeFI	4.51	-0.24	arxiv.org →
15	MoDE + GraspCorrect	4.5	-0.25	arxiv.org →
16	VLA-Adapter (from NS-VLA)	4.5	-0.25	arxiv.org →
17	RoboVLMs (from ELLSA)	4.49	-0.26	arxiv.org →
18	SDP	4.49	-0.26	arxiv.org →
19	HARP-VLA	4.48	-0.27	arxiv.org →
20	Being-H0.7	4.48	-0.27	arxiv.org →
21	FOFPred	4.48	-0.27	arxiv.org →
22	PALM	4.48	-0.27	arxiv.org →
23	NIAF	4.47	-0.28	arxiv.org →
24	VPP + TapSampling	4.46	-0.29	arxiv.org →
25	VLA-IAP (70%) + DreamVLA	4.45	-0.30	arxiv.org →
26	DFM-VLA+Embed	4.44	-0.31	arxiv.org →
27	DreamVLA	4.44	-0.31	arxiv.org →
28	FLOWER (from ThinkProprio)	4.44	-0.31	arxiv.org →
29	VLA-IAP (50%) + DreamVLA	4.44	-0.31	arxiv.org →

Benchmark Leaderboards

Loading leaderboards…