IT之家 9 月 30 日消息，蚂蚁百灵今日发布 Ling-3.1-flash 模型，具备约 560B 总参数，每个 Token 激活约 25B 参数；模型上下文窗口上限为 1M，可容纳更长的文档、代码和任务历史。
官方表示，在研发过程中，围绕通用智能体、搜索、日常办公和软件研发等任务持续优化，并在医疗、金融和材料科学研究等情形持续提高模型能力。
另外，Ling-3.1-flash 将开放为期两周的免费感受；考虑到综合服务成本，免费感受期间模型服务长度为 256K。
免费感受期结束并转为付费服务后，计划开放 1M 上下文。届时，团队也计划同步开源，并继续更新模型表现。
IT之家附这款模型在多个任务实测中得分如下：
本文来源：https://www.ithome.com/1/008/907.htm
若有版权问题，请来信告知：13951431913#139.com 或在右方提交评论。
IT Home News, September 30: Ant Bailing today released the Ling-3.1-flash model, featuring approximately 560B total parameters with about 25B activated per Token; the model's maximum context window is 1M, accommodating longer documents, code, and task histories.
The company stated that during R&D, it continuously optimized for tasks such as general-purpose agents, search, daily office work, and software development, while steadily improving model capabilities in scenarios including healthcare, finance, and materials science research.
In addition, Ling-3.1-flash will offer a free trial for two weeks; considering overall service costs, the model service length during the free trial period will be 256K.
After the free trial period ends and the service transitions to paid, the plan is to open up 1M context. At that time, the team also plans to open-source the model in tandem and continue updating its performance.
IT Home appends the model's scores in real-world tests across multiple tasks as follows: