88Z AI, 16 November — Meituan has open-sourced LongCat-2.5, a 1.6-trillion-parameter agent model trained entirely on Chinese hardware. It scores 88.4 on Terminal Bench, nearing Claude Opus 5, with sample deployment running on 194 Huawei Ascend chips and no Nvidia GPUs involved.
At a glance
- Meituan open-sources LongCat-2.5, a 1.6-trillion-parameter model for AI agents
- Trained and deployed entirely on Huawei Ascend chips without Nvidia GPUs
- Scores 88.4 on Terminal Bench, performing close to Claude Opus 5

Model Release and Hardware Breakthrough
Meituan has officially released the weights for LongCat-2.5, a massive model designed specifically for AI agents with 1.6 trillion parameters. A key breakthrough is that its training relied entirely on domestic Chinese hardware, with no mention of Nvidia GPUs, marking significant progress in de-Americanized computing infrastructure.
Performance and Deployment
On the Terminal Bench benchmark, LongCat-2.5 scored 88.4 points, placing it just behind Claude Opus 5 and demonstrating strong capabilities in agent tasks. The current sample deployment runs on 194 Huawei Ascend chips, providing the industry with a concrete reference path for deploying ultra-large models outside the Nvidia ecosystem.
Sources & editorial note
This article is an AI-assisted synthesis by 88Z AI based on the public sources below. Product claims and availability are attributed to their publishers. It does not represent independent testing or first-hand reporting.
- the-decoder.comSource published
Source dates above refer to the original announcements; the date under the headline is this article’s publication on 88Z AI.