克雷西 发自 凹非寺 量子位 | 公众号 QbitAI 又有一个国产模型，悄悄跻身到了开源第一梯队。 这次不是DeepSeek也不是Qwen，而是小米刚刚官宣的开源模型 MiMo-V2-Flash 。 仅用了309B的参数规模，该模型就展现出了极高的效能密度，在多项权威综合评测中均取得了令人瞩目的优异成绩。 不仅分数高，它还在实现2.6倍推理加速的同时，兼顾了顶尖的模型效果与极致的部署成本。 在小米刚刚举行的"人车家全生态"合作伙伴大会上，小米将该模型定义成了"迈向Agent时代的全新语言基座"。 这个模型在海外也受到了广泛好评，X网友评价说MiMo-V2-Flash将能够让智能体变得更加实用。 还有人在线许愿，希望能推出gguf格式，方便适配自己使用的模型框架。 从技术报告中，我们也了解到了小米在MiMo-V2-Flash背后采用的一系列关键技术： 具体来看—— 给学生模型请一个"私教天团" MiMo-V2-Flash采用了MoE架构，总参数量为309B，包含256个专家，相比那些动辄参数量以T计的巨头模型和2倍参数量的开源模型，可谓 是以小博大。 MiMo-V2-Flash采用了动态激活机制，激活专家数为 ...

Group 1 - The core viewpoint of the news is that Zhifang has successfully completed a new round of Series A financing led by Shenzhen Capital Group, with significant participation from existing shareholders and industry players [1] - The funding will be utilized for the continuous iteration of the GOVLA large model and the AlphaBot series robots, as well as for expanding production lines and global market development [1] - Zhifang has completed seven rounds of financing totaling over 1 billion yuan within six months, indicating strong investor confidence and interest in the company's growth potential [1] Group 2 - Zhifang has launched three generations of robot products, including Alpha Bot, Alpha Bot 1S, and AlphaBot 2, with the latest model featuring 34 degrees of freedom and the ability to perform various tasks without training [2] - The company introduced the world's first full-body VLA large model (GOVLA) in April, which integrates output for full-body control and movement trajectories, enhancing efficiency in task execution [2] - Zhifang has successfully deployed its technology across multiple sectors, including industrial, biotechnology, and public services, while maintaining a focus on creating a self-reinforcing growth path through intelligent-driven commercial models [2]