← Sinan Lab › Sinan Rankings · Robo
Sinan Rankings · measured boards · updated with the index
Sinan Rankings · Robo
Only fields with a source get ranked: release date, parameter count from the model card, pretraining hours as published, declared target bodies, licence as written. No composite score, no recommendation.
Newest releases
By release month on the model card or README
- 01InternVLA-A1.5上海人工智能实验室 InternRobotics2026-072.69 B params
- 02LingBot-VLA 2.0蚂蚁灵波科技 (Ant Group)2026-076.00 B params
- 03Galaxea G0.5星海图(Galaxea)2026-062.00 B params
- 04Isaac GR00T N1.7NVIDIA2026-043.00 B params
- 05Spirit v1.5千寻智能(Spirit AI)2026-015.40 B params
- 06LingBot-VLA 4B蚂蚁灵波(Robbyant)2026-014.20 B params
- 07GR00T N1.6 3BNVIDIA2026 (待核实)3.29 B params
- 08VLA-0NVIDIA Research2025-103.00 B params
Most pretraining data
Published real-robot / human pretraining hours; only models that state a number
- 01LingBot-VLA 2.0蚂蚁灵波科技 (Ant Group)60,000 h
- 02LingBot-VLA 4B蚂蚁灵波(Robbyant)20,000 h
- 03RDT2-VQ清华大学 TSAIL(thu-ml)10,000 h
- 04Galaxea G0.5星海图(Galaxea)500 h
Broadest declared coverage
Number of body forms the vendor declares as targets (declaration, not verified compatibility)
- 01LingBot-VLA 2.0humanoid · dual_arm · single_arm · wheeled4 body forms
- 02π0.5single_arm · dual_arm · mobile_manipulator3 body forms
- 03UniVLAsingle_arm · dual_arm · wheeled3 body forms
- 04GR00T N1.5 3Bhumanoid · dual_arm · single_arm3 body forms
- 05π0 (pi-zero)single_arm · dual_arm · mobile_manipulator3 body forms
- 06RDT-1Bsingle_arm · dual_arm · mobile_manipulator3 body forms
- 07InternVLA-A1.5single_arm · dual_arm2 body forms
- 08Galaxea G0.5dual_arm_wheeled · single_arm2 body forms
Smallest with a commercial licence
Parameter count from the model card, licence clearly allows commercial use
- 01OctoMIT(代码与权重仓库标注)93M B params
- 02X-VLAApache-2.0(代码与模型卡)880M B params
- 03RDT-1BMIT(代码、权重、数据均为 MIT,README 原文)1.00 B params
- 04Isaac GR00T N1.7NVIDIA Open Model License(权重);Apache-2.03.00 B params
- 05EO-1MIT(模型卡)3.77 B params
- 06SpatialVLAMIT(代码 README 与模型卡均标 MIT)4.03 B params
- 07Spirit v1.5代码 MIT(README);模型卡标 apache-2.05.40 B params
- 08LingBot-VLA 2.0Apache-2.06.00 B params
Largest open weights
Parameter count from safetensors metadata where available
- 01Magma-8BMicrosoft Research8.90 B params
- 02RDT2-VQ清华大学 TSAIL(thu-ml)8.29 B params
- 03MolmoActAllen Institute for AI(Ai2)8.12 B params
- 04Being-H0BeingBeyond(北京)8.11 B params
- 05OpenVLAStanford / UC Berkeley / TRI 等(OpenVLA 团队)7.54 B params
- 06UniVLAOpenDriveLab(香港大学 等)7.54 B params
- 07RynnVLA-001阿里巴巴达摩院7.01 B params
- 08LAPAKAIST / 华盛顿大学 等(Latent Action Pretraining)7.00 B params
Most-targeted bodies
How many indexed models declare this body form as a target
- 01松灵 Piper 机械臂松灵机器人 · single_arm31 models
- 02SO-101开源社区 (LeRobot) · single_arm31 models
- 03Franka FR3 / PandaFranka Robotics · single_arm31 models
- 04WidowX 250 STrossen Robotics · single_arm31 models
- 05Google Robot(Everyday Robots)Google / Everyday Robots · single_arm31 models
- 06Franka Emika PandaFranka Robotics · single_arm31 models
- 07智元 精灵 G1智元机器人 · dual_arm_wheeled19 models
- 08Galaxea R1 Pro星海图(Galaxea) · dual_arm_wheeled19 models
Licence at a glance
Counted from licence text as written; when code and weight licences disagree we count it as unverified
Sorted by recorded facts; not a recommendation. Every field links to its evidence on the model page. Data 2026-09-15.
Clear direction, grounded judgement.