Skip to main content

Meituan Denies Rumors: LongCat Model Lead Not Leaving, Stays to Push AI Boundaries

Meituan Denies Rumors: LongCat Model Lead Not Leaving, Stays to Push AI Boundaries

In the fast-paced world of AI, rumors can spread like wildfire. But Meituan is pouring cold water on one such rumor: the reported departure of Pei Peng, the head of its LongCat foundation model team. The company officially stated on July 27 that the information is false, putting an end to speculation.

Who is Pei Peng?

Pei Peng is a key figure in Meituan's AI ambitions. A Peking University graduate, he joined Meituan in 2023 and quickly rose through the ranks. He served as pre-training lead for the base model team before taking the helm of the LongCat team's foundation model division. His role? To lead the research and development of Meituan's large language models, multimodal models, and Agent intelligent models. In short, he's the brain behind Meituan's push into cutting-edge AI.

LongCat-2.0: A Milestone in Domestic Computing

Meituan's LongCat-2.0, released on June 30, is a game-changer. It's the first trillion-parameter large model in the industry to complete training and inference using domestic computing power throughout the entire process. That's a big deal for China's tech independence. The model uses a Mixture of Experts (MoE) architecture, with a total of 1.6 trillion parameters. For each token, about 48 billion parameters are activated, making it both powerful and efficient. It also supports a super-long context of 1 million tokens and is deeply compatible with mainstream harnesses like Claude Code.

Why This Matters

Meituan's quick clarification of the personnel rumors shows its strategic commitment to developing self-controlled large models. The stability of its core team is crucial as the company continues to push the boundaries of what's possible with domestic AI technology. By denying the rumors, Meituan is signaling that it's all in on LongCat and the future of AI.

Key Points

  • Rumor Denied: Meituan officially denies that Pei Peng, head of the LongCat foundation model team, is leaving.
  • Pei Peng's Role: A Peking University graduate, Pei Peng leads the development of Meituan's large language models, multimodal models, and Agent intelligent models.
  • LongCat-2.0: Released on June 30, it's the first trillion-parameter model trained and inferred entirely on domestic computing power.
  • Technical Specs: Uses MoE architecture with 1.6 trillion total parameters, 48 billion activated per token, and supports 1M token context.
  • Strategic Commitment: Meituan's quick response underscores its dedication to domestic AI development and team stability.