Physix Frontier · News Briefing Card (SSPAI · Nov 3, 2026)

Meituan Releases 1.6T-Parameter LongCat-2.5 Multimodal Model

KEY FACTS

  • Meituan's LongCat team has released LongCat-2.5, a natively multimodal model with 1.6 trillion total parameters.
  • The model activates roughly 48 billion parameters per token, strengthening visual understanding and agentic coding.
  • Training incorporates long-horizon trajectory data and large-scale long-horizon reinforcement learning to improve end-to-end engineering capability.
  • The model converts visual signals into multi-level discrete tokens and processes them uniformly with text.
  • Moma launched the Tanjing 3 livestreaming camera, priced at 3199 yuan, supporting 4K 60p livestreaming.

KEY DATA

1.6 trillionTotal model parameters
About 48 billionActivated parameters per token
3199 yuanTanjing 3 price
50 millionTanjing 3 effective pixels

PHYSIX OBSERVATION

Meituan is betting its trillion-scale multimodal model on long-horizon agentic tasks, aiming at real engineering scenarios rather than chat demos - a signal that large-model competition is shifting toward deployment capability. But parameter scale does not equal usability; success rates on long-horizon tasks and inference cost are what users will actually pay for.

Source: SSPAI report