# Meituan Opens LongCat-2.5 Model Weights Under MIT License

By Simon Yoon

Canonical URL: https://www.tokenpost.com/news/technology/57779
Published: 2026-11-15T11:39:11.000Z
Updated: 2026-11-15T11:39:11.000Z
Section: Technology

> The self-hosting guide is built around 194 Huawei Ascend chips. The model supports a context window of up to one million tokens.

Meituan’s LongCat team released its 1.6 trillion-parameter LongCat-2.5 weights under the MIT License, with a self-hosting guide built around 194 Huawei Ascend chips for developers who want to run the model themselves.

The model uses a mixture-of-experts design, with about 48 billion parameters active for each token, and supports a context window of up to one million tokens. Meituan also released a 4-bit quantized version. The [earlier launch of LongCat-2.5](<https://www.tokenpost.com/news/technology/47537>) included visual processing and agent features for extended software tasks.

In Meituan’s comparison, LongCat-2.5 scored 88.4 on Terminal Bench v2.1, compared with 89.1 for Claude Opus 5 and 88.8 for GPT-5.6 Sol.

Meituan’s limited-time API pricing is about €0.27 per million input tokens and €1.07 per million output tokens.



## Links in this article

- [earlier launch of LongCat-2.5](https://www.tokenpost.com/news/technology/47537)