Meituan leads Series A investment, Mindverse raises nearly $50 million in total funding, and will open source the first 750B reinforcement learning agent model based on GLM 5.1 According to Beating monitoring, AI intelligent agent developer Mindverse (Xinzhou Technology) announced the completion of Series A financing led by Meituan, with joint investments from Yuanhe Puhua, Shaoyin, Variable Capital, and existing shareholders, raising nearly $50 million in total. Gaohu Capital serves as the exclusive financial advisor. Mindverse was co founded by Chen Kaijie and Andrew (who previously collaborated with Yao Shunyu to publish ReAct/FireAct papers), with a core R&D team from DeepSeek, ByteSeed, xAI, and others, dedicated to building native agent models through post training and continuous learning techniques. Unlike the conventional approach of relying on external Prompt concatenation, Mindverse firmly believes that the core capability of the agent comes from the "Post Training" of the model. The company uses LoRA as the technological foundation for building continuous learning and parameterized memory, and has achieved lossless LoRA reinforcement learning for trillion parameter MoE models by the end of 2025, reducing post training costs to one tenth of full parameter training. Through the Mixture of LoRA architecture, thousands of independent "skill packs" (LoRAs) representing different preferences or business experiences can be mounted on a shared base, supporting dynamic fast activation (in seconds) and clean privacy isolation. Mindverse announced that it will open source its latest trained 750B parameter agent model, which includes a 744B pre trained GLM 5.1 base and a 6B LoRA. It is the world's first achievement to complete reinforcement learning training on GLM 5.1 and has achieved SOTA in the Living Bench, Vita Bench, A2UI, and PinchBench rankings. On the product side, Mindverse adopts a bidirectional flywheel that combines model training with the C-end life agent application Macaron (2 million users, 100000 daily active users) for collaborative iteration; Commercially, we have collaborated deeply with Huawei Cloud, Microsoft Cloud, and other companies to open up low-cost and high concurrency LoRA training and inference infrastructure to enterprises. [Original link]