ACE-Brain-0.6
ACE Robotics, founded by SenseTime co-founder Xiaogang Wang, released ACE-Brain-0.6, a spatial grounding model for robots that beat GPT-6 Astra on six of eight grounding benchmarks.
ACE-Brain-0.6 is built by supervised fine-tuning of Qwen3.5-9B on about 1 million spatial-grounding samples. It identifies objects, locates them in images and works out where a robot can interact with them. The company reports an average of 74.53 across seven public benchmarks and 94.13 on RobotGrounding, against 73.68 and 91.10 for GPT-6 Astra.
- Date
- Monday 19 October 2026
- Lab
- ACE Robotics
- Kind
- model
- Access
- paper only
Figures
| Measure | Value | Measured by |
|---|---|---|
| Average score across seven public grounding benchmarks | 74.53 GPT-6 Astra scored 73.68 | company |
| RobotGrounding score | 94.13 Real robot scenario benchmark; GPT-6 Astra scored 91.10 | company |
| Grounding benchmarks outperforming GPT-6 Astra | 6 of 8 | company |
| Training samples | about 1 million spatial-grounding samples | company |
Access is not stated in the source, so the access value is a placeholder; benchmark results are company-reported and were not independently confirmed.
Sources
This record was partly confirmed: some claims could not be checked on 20 October 2026. How we check