LingBot-VA 2.0
LingBot-VA 2.0 is a video-action foundation model trained from scratch for robot control. It combines semantic visual-action tokenization, causal pretraining, a sparse mixture-of-experts backbone and asynchronous inference grounded in current observations.
Access and availability
SourceVerified on · 2026-10-30
At a glance
- Released
- July 9, 2026
Model card
Section 4.1.1 reports approximately 15.3 billion trained parameters and 2.5 billion active parameters per inference token, including the video backbone, action expert and multi-chunk prediction heads. Version-specific weights and license remain unverified; base weights belong to the earlier version. https://arxiv.org/html/2607.08639v2#S4.SS1.SSS1 https://technology.robbyant.com/lingbot-va-v2
Before you use this version: Available weights or code do not establish plug-and-play robot compatibility. Check the required configuration, data and interfaces in the source documentation.