As global embodied intelligence competition shifts from hardware bodies to "brains," Unitree Robotics is advancing its position with a new general-purpose humanoid robot foundation model, UnifoLM-WLA-1.0. The model has 6 billion parameters and was trained on roughly 2,500 hours of real robot data.
The model claims top open-source performance on seven embodied reasoning benchmarks, with several spatial understanding tasks exceeding GPT-6 Astra.
The embodied reasoning foundation is UnifoLM-ER-1-4B, trained on Qwen3-VL-4B using more than 5 million embodied reasoning samples. It leads the compared open-source models on seven of 16 multimodal perception and understanding benchmarks. On Where2Place it scored 82.0 against GPT-6 Astra's 69.0, and on EmbSpatial the two scored 88.9 and 83.3 respectively.
For robots, spatial understanding matters more than conversational ability, since knowing where an object is, how it should be positioned and what happens next is a precondition for acting correctly and for operating in the real world.
A single UnifoLM-WLA-1.0 model can coordinate 64 tasks, ranging from folding towels and tidying dishes to making a bed, taking out trash and putting clothes into a washing machine, demonstrating broad generalization.
This addresses a long-standing industry problem. Past demos were typically single-task, single-scenario and highly customized, presenting well but failing to transfer when the environment changed, making them difficult to scale. Whether one model can cover more tasks and scenarios is an important step toward general-purpose robots.
Unitree announced it will open-source the model, code and dataset. The move positions Chinese manufacturers as participants in defining the "embodied intelligence brain" layer, as the global contest moves from body specifications to general-purpose model foundations. Competitors include Google's Gemini Robotics 2, which promotes one brain for any robot, and comments from Nvidia's Jim Fan arguing that VLA has had its day and world action models should prevail.
Source: Anhui Broadcasting / Securities Times
