TARS unveils AWE 3.5, a generalist embodied AI model

Robotics firm TARS released AWE 3.5, an embodied foundation model trained on a million hours of data that handles diverse tasks like phone packing, screw sorting and cable management. Its embodied-native architecture combines latent actions and geometric priors, aiming for generalization rather than single-task execution.

2026-08-18 ~ 2026-08-20 · 3 related posts

• TARS AWE 3.5 aims for generalization with embodied-native architecture — FellMentKE · 2026-08-18
• TARS AWE 3.5 Shows Cross-Scene Generalization with Million-Hour Training — heyshrutimishra · 2026-08-19
• TARS Unveils AWE 3.5 General-Purpose Robot Model — heyshrutimishra · 2026-08-20