HKU-Led Embodied AI Benchmark: Zhiyuan Model Tops, Surpassing General LLMs
On October 27, the University of Hong Kong-led embodied AI evaluation benchmark released its latest leaderboard. The benchmark assesses performance across various embodied tasks. Zhiyuan's model secured the top position with an average score of 61.90 and an average success rate of 56.42%, outperforming general-purpose large models.
The results highlight a trend toward division of labor and collaboration between general large models and embodied specialized models. While general models excel at broad reasoning, embodied models are optimized for physical interaction and task execution.
The establishment of this benchmark provides the industry with a relatively objective capability metric. It enables fair comparison across different embodied AI systems and helps guide future development.