Code & Chain · Signal Desk

Meituan releases 1.6-trillion-parameter LongCat-2.5, trained entirely without Nvidia GPUs

Original sourceThe Decoder

Summary

Meituan has released LongCat-2.5, a 1.6-trillion-parameter agent model trained entirely on Chinese hardware with no Nvidia GPUs. It uses a mixture-of-experts architecture activating only about 48 billion parameters per token, supports prompts of up to 1 million tokens and responses of up to 128k to…

Key points

  • Gives readers a handle on hardware autonomy and open-source licensing options for frontier agent models, with direct implications for self-hosting and cost planning.
  • This is the first time a large frontier agent model has been trained on non-Nvidia hardware and released as open source, potentially reshaping model supply and compute competition.
  • Enterprises and developers can assess the feasibility of self-hosting on Ascend clusters; investors should watch the long-term impact of Chinese compute autonomy on GPU supply-chain expectations.

Editorial note

This page is Code & Chain's editorial summary of public sources. It may be prepared with AI assistance and published through an automated workflow. Refer to the original sources; this content is not investment, legal, or tax advice.