Meituan's LongCat Platform Releases LongCat-2.5-Preview Model with 1.6T Parameters

On September 25, Meituan's LongCat API platform launched LongCat-2.5-Preview, a new large language model focused on long-horizon tasks and multimodal capabilities, with API and web access available immediately.

According to the official changelog, LongCat-2.5-Preview introduces image understanding, enabling cross-modal Q&A, content summarization, and complex visual reasoning. The model also demonstrates strong performance in code generation, code comprehension, and automated programming tasks, with deep integration support for Claude Code, Hermes, OpenClaw, OpenCode, and Kilo Code.

The model follows a Mixture-of-Experts (MoE) architecture with approximately 1.6 trillion total parameters and about 48 billion activated parameters per inference. It natively supports a 1 million token context window, suited for processing extremely long documents, codebases, logs, and multi-turn complex tasks.