Skip to main content
ModelsResearch

Meituan releases a 1.6-trillion-parameter agent model that never touched an Nvidia GPU

Source: The Decoder (opens in a new tab) · Jonathan Kemper

Intel Summary

Meituan has released the weights for LongCat-2.5, a 1.6-trillion-parameter model designed for AI agents and trained entirely on Chinese hardware without Nvidia GPUs. The model reportedly achieved 88.4 points on Terminal Bench, ranking closely behind Claude Opus 5, with sample deployment configurations running on 194 Huawei Ascend chips.

Why It Matters

The release indicates that trillion-parameter agent models can be trained and deployed using non-Nvidia hardware ecosystems such as Huawei Ascend processors. This provides a reference point for organizations evaluating open-weight agent models and non-Western hardware infrastructure for frontier-scale model development.

Organizations & Entities

Topics