# Meituan Launches New Large Model LongCat-2.5-Preview

- Category: Flash (快讯)
- Published: 2026-09-28T11:58:00.000Z
- Source date: 2026-09-28
- HTML: https://www.88z.ai/en/news/flash-20260928-833338e909b17934
- Markdown: https://www.88z.ai/en/news/flash-20260928-833338e909b17934.md
- Other language: https://www.88z.ai/news/flash-20260928-833338e909b17934.md

Meituan has released the new LongCat-2.5-Preview large model, maintaining a MoE architecture with 1.6T total parameters and approximately 48B activated parameters. It supports a 1M token ultra-long context window and a maximum output of 128K tokens. The model natively integrates image understanding capabilities, enabling cross-modal Q&A and visual reasoning, and can execute long-process Agent tasks for terminals, browsers, and desktop software.

## At a glance

- Maintains 1.6T parameter MoE architecture with 1M context and 128K output
- Natively integrates image understanding for cross-modal reasoning and long-process Agent tasks

## The update

Meituan has officially launched the new generation large model, LongCat-2.5-Preview. The model continues the previous architectural design, featuring a Mixture of Experts (MoE) structure with 1.6T total parameters and approximately 48B activated parameters. It maintains the capability to process an ultra-long context of 1M tokens, with a maximum output length of up to 128K tokens. The core highlight of this upgrade is the native integration of image understanding capabilities into the base model, enabling more efficient cross-modal question answering and visual reasoning. With these enhanced features, LongCat-2.5-Preview can execute complex long-process Agent tasks for terminal devices, browsers, and desktop software, further expanding the operational capabilities and application scope of large models in practical scenarios.

## Sources

- [LongCat-2.5-Preview](https://www.88z.ai/flash/cite/flash-20260928-833338e909b17934) (2026-09-28)

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.
