Mindverse announced the completion of its Series A funding round led by Meituan, raising nearly $50 million. Participants in this round included Yuanhe Puhua, Shokz, Bianliang Capital, and other followers, with existing investors also increasing their stakes. Mindverse boasts an impressive roster of historical investors, including top-tier funds such as Ant Group, Yuanyi Capital, Sequoia China, ZhenFund, and GaoRong Capital. GaoHu Capital served as the exclusive financial advisor for this funding round. Mindverse focuses on building advanced Agent models capable of continuous learning. Its core approach involves utilizing LoRA technology to attach lightweight "skill packages" to general large models, enabling low-cost accumulation of memories and capabilities specific to users or scenarios. This approach avoids the high costs associated with retraining from scratch each time. The company believes that an Agent's capabilities should stem from training rather than simple prompt engineering and framework assembly. The Mindverse team, comprising around 20 core R&D members with impressive backgrounds, has collectively published over 200 papers at top conferences. Founder Kaijie Chen and others, drawing from their entrepreneurial experiences, firmly believe that capabilities for long-term tasks must be acquired through training. The company's soon-to-be-open-sourced 750B agent model represents the world's first to undergo reinforcement learning post-training on GLM 5.1. This model supports mixture of LoRA and is optimized for agent scenarios, serving purposes such as generative UI coding, chatting, long-chain reasoning, and tool invocation. Through Macaron as its agent harness, Mindverse has established a bidirectional loop between model training and product iteration. The company prioritizes user privacy by training models in simulated environments. Currently, Macaron boasts over 2 million users with daily active users exceeding 100,000. In terms of commercialization, Mindverse collaborates with companies like NVIDIA, ByteDance, and Ant Group on technology co-development and framework partnerships. It also works with cloud providers like Huawei Cloud to reduce training costs for clients. The company itself remains focused on technological R&D, with its C-end products centered around Macaron. Mindverse emphasizes the importance of isolation, viewing LoRA as an effective means of achieving memory segregation. It asserts that true long-term memory should be trained into the model parameters rather than relying on external notebooks. The company also explores memory systems like δ-mem and positions itself as a post-training company that derives model requirements from real-world scenarios, fitting the Neo Lab mold. Its goal is to create a flywheel effect between agent models and C-end products, prioritizing research breakthroughs with the belief that technological advancements will drive commercial value. Ultimately, Mindverse aspires to enable AI to understand each unique individual and reduce user frustrations.