This Double 11, Bailing's new model works without burning out

What happened: Ant's Bailing released Ling-3.1-flash, an open source model built for real workflow tasks, and it has been integrated into Ant's internal desktop agent client Xiaohu. Key numbers: About 560B total parameters, about 25B activated per token. Context window up to 1M. 512 routed experts, 8 selected per token plus 1 shared expert. Why it matters: Bailing positions the model around task delivery rather than benchmark scores, and the content says its efficiency framing has drawn industry response as Anthropic and OpenAI add finance and medical capabilities.

Ant's Bailing released Ling-3.1-flash, an open source model built for real workflow tasks, and it has been integrated into Ant's internal desktop agent client Xiaohu.

Bailing positions the model around task delivery rather than benchmark scores, and the content says its efficiency framing has drawn industry response as Anthropic and OpenAI add finance and medical capabilities.