IT之家 · 智能时代T2·68 pts
- ACE-Brain-0.6 is supervised fine-tuned on Qwen3.5-9B using approximately one million spatial anchoring training samples.
- Outperforms OpenAI GPT-6 Astra in six of eight grounding benchmarks, with an average public benchmark score of 74.53 vs 73.68.
- Achieves a score of 94.13 in the real-world RobotGrounding evaluation, significantly higher than Astra's 91.10.