Daxia Robotics has introduced its new embodied foundation model, ACE-Brain-0.6. The model is built on a supervised fine-tuning of Qwen3.5-9B and is designed to improve target recognition, spatial positioning, and interaction area assessment.
The development team utilized approximately one million spatial anchoring training samples for the model. In a series of eight Grounding evaluations, ACE-Brain-0.6 outperformed OpenAI GPT-6 Astra in six categories. The model achieved an average score of 74.53 across seven public benchmarks, compared to 73.68 for the competitor.
In the RobotGrounding evaluation, which focuses on real-world robotic scenarios, the model scored 94.13 points, surpassing the 91.10 points achieved by Astra. These results highlight the model's capabilities in physical AI applications.
Daxia Robotics was founded by Wang Xiaogang, a co-founder of SenseTime. The company focuses on embodied intelligence and utilizes its proprietary ACE research paradigm and the Kaiwu world model. As of October 2026, the company has secured several hundred million USD in funding.
