88Z AI, 10 July — Ant Lingbo Tech has released LingBot-VA 2.0, the industry's first embodied-native world action model. Built on a zero-based pre-trained autoregressive architecture, it integrates four core designs: semantic visual-action tokenizer, causal pre-training, MoE architecture, and asynchronous inference. It supports 150Hz real-time inference on a single card, enabling robots with general control capabilities to reason and act simultaneously.
At a glance
- LingBot-VA 2.0 is the industry's first embodied-native world action model, pre-trained from scratch on an autoregressive architecture.
- It achieves 150Hz real-time inference on a single card via four core designs, enabling robots to reason and act simultaneously.
The update
Ant Lingbo Tech has officially launched LingBot-VA 2.0, the industry's first embodied-native world action model. Pre-trained from scratch using an autoregressive architecture, the model aims to enhance general robotic control. Its technical core features four key designs: a semantic visual-action tokenizer, a causal pre-training paradigm, a Mixture of Experts (MoE) architecture, and an asynchronous inference mechanism. These innovations enable the model to achieve real-time inference speeds of up to 150Hz on a single GPU. This performance breakthrough allows robots to perform logical reasoning and physical actions simultaneously, significantly improving response efficiency and control precision in complex environments, offering a new technical path for embodied intelligence.
Sources & editorial note
This article is an AI-assisted synthesis by 88Z AI based on the public sources below. Product claims and availability are attributed to their publishers. It does not represent independent testing or first-hand reporting.
- 蚂蚁灵波科技Source published
Source dates above refer to the original announcements; the date under the headline is this article’s publication on 88Z AI.