mindlab-research/Macaron-V1-Preview-749B
MindLab Research has released Macaron-V1-Preview-749B, a 749-billion parameter Mixture-of-LoRA model designed for personal agent tasks. This model, built upon GLM-5.1 with MinT, features a 744B base model augmented by five specialist LoRA adapters. It is engineered to handle multi-turn interactions involving user intent, private state, tools, and world state, utilizing an explicit router-tool design to switch between adapters for tasks like personal planning, coding, and generative UI. AI
IMPACT This model's specialized adapters for personal agents, coding, and UI generation could accelerate development of more capable and context-aware AI assistants.