Woofun AI reports that Mindverse, an AI intelligence R&D company, has completed a Series A funding round led by Meituan, raising nearly $50 million. The round saw participation from Yuanhe Bohai, ShawYin, Variant Capital, and existing shareholders, with GaoHu Capital serving as the exclusive financial advisor. Co-founded by Chen Kaijie and Andrew, Mindverse boasts a core R&D team with backgrounds at DeepSeek, Byte Seed, and xAI, focusing on native agent models through post-training and continual learning.

The company differentiates itself by prioritizing 'post-training' over external prompts, utilizing LoRA as the foundation for continuous learning and parameterized memory. By late 2025, Mindverse achieved lossless LoRA reinforcement learning on a trillion-parameter MoE model, reducing post-training costs to one-tenth of full-parameter training. This Mixture of LoRA architecture allows thousands of independent 'skill packs' to be mounted on a shared base, enabling rapid activation and privacy isolation.

Concurrently, Mindverse plans to open-source its 750B parameter Agent model, comprising a 744B pre-trained GLM 5.1 base and 6B LoRA, which holds SOTA rankings in Living Bench, Vita Bench, A2UI, and PinchBench. Commercially, the firm synergizes model training with its C-end lifestyle Agent application Macaron, which serves 2 million users, while partnering with Huawei Cloud and Microsoft Cloud for low-cost inference infrastructure.