Daxiao Robotics Unveils ACE-Brain-0.6 Embodied Model, Beats GPT-6 Astra on Six Spatial Tests

On October 19, Jiemian News reported that Daxiao Robotics released ACE-Brain-0.6, a visually spatial-anchoring enhanced embodied foundation model. The model is built on Qwen3.5-9B and fine-tuned with supervised learning, aiming to strengthen object recognition, spatial localization and interaction-region judgment.

According to Daxiao Robotics, the model was trained on roughly one million spatial-anchoring samples. The company said these samples help anchor visual features to spatial references and task-relevant regions, supporting more reliable perception in physical environments.

In eight grounding evaluations, ACE-Brain-0.6 surpassed OpenAI's GPT-6 Astra on six benchmarks, according to the company. Grounding, the ability to link language to objects and locations, is central to embodied systems that must act in physical spaces. The results indicate that targeted spatial-anchoring fine-tuning can deliver competitive embodied perception at a relatively compact model scale. The company positions the model as an upgrade focused on the perception layer, reflecting growing attention to spatial intelligence in embodied AI. Independent verification and broader task testing will still be needed to assess real-world robustness.