IT之家 9 月 26 日消息,美团旗下 LongCat API 开放平台于 9 月 25 日上线新一代大模型 LongCat-2.5-Preview,主打“长程任务”与多模态能力,并同步开放 API 与网页端感受入口。
IT之家附上官方更新日志内容如下:
LongCat-2.5-Preview 的核心特性如下:
依据官方公布的信息,LongCat-2.5-Preview 模型延续 MoE(混合专家)路线,总参数量约 1.6 万亿,每次推理激活参数约 480 亿。该模型兼容原生兼容 100 万 token 上下文窗口,合适处理超长文档、代码库、日志与多轮复杂任务。
参考
本文来源:https://www.ithome.com/1/007/356.htm
若有版权问题,请来信告知:13951431913#139.com 或在右方提交评论。
Thanks to IT Home netizens GreatMOLA, Domado, and SoftMedia user 389454 for their tip submissions!
IT Home, September 26 - Meituan's LongCat API open platform launched the new-generation large model LongCat-2.5-Preview on September 25, focusing on "long-horizon tasks" and multimodal capabilities, while also opening API and web-based access for experiencing the model.
IT Home provides the official changelog as follows:
The core features of LongCat-2.5-Preview are as follows:
According to officially released information, the LongCat-2.5-Preview model continues the MoE (Mixture of Experts) approach, with a total of approximately 1.6 trillion parameters and approximately 48 billion activated parameters per inference. The model is natively compatible with a 1 million-token context window, suitable for handling ultra-long documents, codebases, logs, and complex multi-turn tasks.
Reference

