Skip to main content

Meituan Quashes Rumors: LongCat AI Lead Stays, Pushing Domestic Computing

Meituan Quashes Rumors: LongCat AI Lead Stays, Pushing Domestic Computing

Meituan has officially denied swirling rumors that Pei Peng, the head of its LongCat foundation model team, is about to jump ship. The company stated unequivocally that the information is false, putting to rest speculation that had been making the rounds.

Pei Peng, a Peking University graduate, joined Meituan in 2023. Since then, he has worn multiple hats: pre-training lead for the base model team, and head of the LongCat team's foundation model. In these roles, he has been the driving force behind Meituan's large language model, multimodal model, and Agent intelligent model. Most notably, he led the development of LongCat-2.0, a trillion-parameter behemoth.

LongCat-2.0, released on June 30, is a milestone. It's the first trillion-parameter large model in the industry to complete both training and inference using domestic computing power from start to finish. The model uses a Mixture-of-Experts (MoE) architecture, boasting a total of 1.6 trillion parameters, with about 48 billion parameters activated per token. It supports a super-long context of 1 million tokens and is deeply compatible with mainstream tools like Claude Code.

By quickly addressing the personnel rumors, Meituan has demonstrated its strategic resolve and the stability of its core team in the race to develop self-controlled large models. The company is clearly betting big on homegrown AI infrastructure.

Key Points

  • Meituan denies rumors that LongCat AI lead Pei Peng is leaving.
  • Pei Peng led the development of LongCat-2.0, a trillion-parameter model.
  • LongCat-2.0 is the first such model trained entirely on domestic computing power.
  • The model uses MoE architecture with 1.6 trillion total parameters.
  • Meituan emphasizes team stability and commitment to domestic AI development.