Embodied AI APIs Move Models Beyond Launch Videos
GeekPark reports that Physical Intelligence opened a model API test, Dream Labs launched Dream API with duration-based inference billing and fine-tuning, and Dexmal expanded Dexmal MaaS with an integrated workflow for data upload, training, deployment, and improvement from real-robot failures. These services let developers connect their own robots, adapt models to specific tasks, and test cloud inference on real-world workloads. The report says cross-embodiment adaptation remains limited by hardware, kinematics, control systems, and tasks, while reliable customer-task completion still requires adaptation and field validation.
The reported launches shift embodied AI from demonstrations toward developer-accessible services, but their practical value depends on post-training and real-robot reliability rather than general API availability.

具身智能的 API 时刻意味着部分具身模型已经开始具备跨本体迁移、在一定范围内适应新任务的能力。
作者|Li Yuan
编辑|郑玄
具身智能,也到了 API 时刻?
过去,想知道一个具身模型有多强,往往要看它在发布视频里完成了什么任务。不过,这条线正在松动。
#Physical Intelligence#Pi API#Dream Labs#Dream API#Dexmal MaaS#DM0.5#DexOS Lite#Almond#Axol