ACE Robotics releases ACE-Brain-0.6, a 9B visual grounding model

ACE Robotics has unveiled ACE-Brain-0.6, a roughly 9-billion-parameter embodied model built on Qwen3.5-9B and trained on about one million spatial supervision samples (940,000 images, 2.7 million QA pairs). The company claims it beat OpenAI's GPT-6 Astra on six of eight grounding benchmarks, with a 74.53 average versus 73.68.

• Built on Qwen3.5-9B with a frozen vision encoder and fine-tuned language model
• Trained on ~1M samples: 940,000 images and 2.7 million question-answer pairs
• Average score of 74.53 vs. 73.68 for GPT-6 Astra across seven benchmarks
• Scaling data from 260K to 1M samples lifted the average from 69.3 to 74.5