UBTECH Thinker-WM

Thinker-WM 1.0 · Family: Thinker

World model from the UBTECH Thinker family: models the physical environment and predicts the outcomes of a robot action. Ranked first on the LIBERO benchmark.

✓ Active✓ Public access⚖ Open sourceWorld Model📁 Thinker

🏢UBTECH RoboticsProducer

Deployment:📱 On-device☁ Cloud

Overview

UBTECH Thinker-WM is a world model from the Thinker family by UBTECH Robotics. It is the second tier of the three-layer Thinker stack and is used to model the physical environment and predict the outcomes of actions before a robot executes them.

Results and roadmap

Thinker-WM ranked first on the LIBERO benchmark for robot-manipulation policies (the exact score was not disclosed). UBTECH announced a next generation, Thinker-WM 2.0, for the second half of 2026. Architecture details and parameter counts have not been publicly disclosed.

Classification

World Model

Family: Thinker

Applications

Robotic manipulation Robot policy training Interactive world simulation

Access & deployment

On-deviceCloud

Weights: Open source

Key parameters

📥 Input: image, video, robot sensors, robot state data

Robotics

Environment modelingSpatial predictionScene understanding3D scene reconstruction

Technical specification

License

Open source (deklaracja UBTECH; brak potwierdzonego publicznego repozytorium)

Hardware requirements

Not publicly disclosed. A world model in the cognitive layer of UBTECH humanoids (on-device and cloud inference).

Modalities

⬇ Input

imagevideorobot_sensorsrobot_state_data

⬆ Output

videomotion_trajectoriesrobot_actions

Capabilities and applications

Native model capabilities

World simulation

Model's ability to generate coherent, interactive simulations of physical environments — maintaining geometry, lighting, and physics during exploration.

Category: multimodal

Planning

Forming and executing action plans for complex tasks.

Category: planning

Multimodal understanding

Category: multimodal

Multi-step reasoning

Carrying out multi-step chains of reasoning across long, complex tasks.

Category: reasoning

Robotics

Environment modelingSpatial predictionScene understanding3D scene reconstruction

Application domains