On October 19, Daxiao Robotics released Ace-brain-0.6, an enhanced physical model with enhanced visual spatial anchoring. The model is supervised and fine-tuned based on Qwen3.5-9B, focusing on improving target recognition, spatial positioning, and interactive area judgment capabilities. According to Daxiao's disclosure, the model used about one million spatially anchored training samples, and the results surpassed OpenAI GPT-6 Astra in six of the eight Grounding evaluations. Among them, the average score for the seven public benchmarks was 74.53, higher than the latter's 73.68; 94.13 points were obtained in the self-built RobotGrounding evaluation.

On October 19, Daxiao Robotics released Ace-brain-0.6, an enhanced physical model with enhanced visual spatial anchoring. The model is supervised and fine-tuned based on Qwen3.5-9B, focusing on improving target recognition, spatial positioning, and interactive area judgment capabilities. According to Daxiao's disclosure, the model used about one million spatially anchored training samples, and the results surpassed OpenAI GPT-6 Astra in six of the eight Grounding evaluations. Among them, the average score for the seven public benchmarks was 74.53, higher than the latter's 73.68; 94.13 points were obtained in the self-built RobotGrounding evaluation.