Embodied AI Firm Daxiao Robotics Releases ACE-Brain-0.6, Claims Superior Spatial Grounding vs. GPT-6 Astra
Daxiao Robotics released the ACE-Brain-0.6 embodied model based on Qwen3.5-9B, disclosing self-tested scores that surpass OpenAI's GPT-6 Astra in several spatial grounding benchmarks.
ImportanceLocalEvidenceE2 unreplicatedWrite-upQuick
Chinese embodied AI company Daxiao Robotics released ACE-Brain-0.6, a vision-spatial anchoring enhanced embodied foundation model, on October 19, 2026. Founded by SenseTime co-founder Wang Xiaogang, Daxiao developed the model using supervised fine-tuning on Qwen3.5-9B to improve target recognition, spatial localization, and interaction area judgment for physical AI applications.
According to Daxiao's disclosure, ACE-Brain-0.6 was trained on approximately one million spatial anchoring samples. The company claims that in eight Grounding evaluations, six results exceeded those of OpenAI's GPT-6 Astra. Specifically, across seven public benchmarks, ACE-Brain-0.6 achieved an average score of 74.53, compared to 73.68 for GPT-6 Astra. In the RobotGrounding evaluation for real-world robot scenarios, it scored 94.13, surpassing Astra's 91.10. The announcement did not specify whether tests were simulated or on real hardware, nor did it disclose specific hardware configurations.
Daxiao Robotics, considered a key new business within the SenseTime ecosystem for embodied intelligence, has completed hundreds of millions of dollars in funding as of October 2026.