AI 导读
蚂蚁百灵 Sante 强化深度研究与循证检索能力,在启用工具时 HLE 得分 53.24,采用多智能体方案后在 BrowseComp 上达到 86.89。团队还探索了多智能体协作在检索密集型任务中的价值,在复杂搜索与信息获取任务上表现强劲。
正文
Retrieval matters enormously in healthcare. Many real-world tasks depend on finding the right information and grounding decisions in reliable evidence, so strengthening Sante’s deep research and evidence-based retrieval capabilities was a key focus.
We also explored the value of multi-agent collaboration for retrieval-intensive tasks. Sante achieves 53.24 on HLE with tools and reaches 86.89 on BrowseComp with a multi-agent setup, showing strong performance across complex search and information-seeking tasks.
来源:@AntLingAGI · x.com