Meituan has officially launched and made available its advanced AI model, LongCat-Flash-Chat, on August 31, 2025. This significant release, accessible across platforms like GitHub, Hugging Face, and Meituan's official website, introduces a powerful 560 billion parameter Mixture-of-Experts (MoE) architecture designed for enhanced efficiency and performance, particularly in agent-based tasks.
The LongCat-Flash-Chat model represents a sophisticated approach to managing large-scale AI, employing an MoE architecture that activates only a fraction of its parameters—between 18.6 to 31.3 billion—per context. This selective activation, often referred to as sparse activation, drastically reduces computational overhead and allows for faster inference speeds compared to traditional dense models where all parameters are engaged for every task. This efficiency is crucial in today's rapidly evolving AI landscape, where balancing model capacity with computational cost is paramount. The adoption of MoE architecture is a growing trend, with other leading models like Mistral's Mixtral 8x7B and potentially OpenAI's GPT-4 also utilizing this approach to manage massive parameter counts effectively.
Meituan's foray into advanced AI development is not new, following their earlier launch of the AI Coding Agent, NoCode, in June 2025. This sustained focus signals a strategic commitment to pushing the boundaries of AI research and application within the company. The open-sourcing of LongCat-Flash-Chat democratizes access to cutting-edge AI technology, fostering innovation and enabling a wider community of developers to build upon Meituan's architectural advancements.
Open-source AI models are increasingly valued for their cost-efficiency, accessibility, and customization potential, with many organizations, particularly in the technology sector, reporting increased adoption and satisfaction with these models. The development and release of LongCat-Flash-Chat align with broader industry trends where open-source AI models are rapidly closing the performance gap with proprietary systems. This democratization of AI tools allows businesses of all sizes to leverage advanced capabilities without the prohibitive costs often associated with closed-source solutions. The emphasis on agentic behaviors and faster inference speeds suggests Meituan is targeting practical applications where real-time processing and intelligent task execution are critical. The model's architecture, which optimizes computation-communication overlap, further contributes to its efficiency, enabling high throughput and low latency inference, even on large-scale deployments. This move by Meituan underscores the growing importance of open collaboration and shared innovation in advancing the field of artificial intelligence, offering new avenues for growth and development across various sectors.