Meituan: LongCat 2.0

Meituan: LongCat 2.0 is a text model from Meituan, released by the provider on 2026-07-20. It is a sparse mixture-of-experts language model, with 48B active parameters out of 1.6T total. The model id on api14 is meituan/longcat-2.0, and its main category is text.

LongCat 2.0 accepts text input and returns text output. It has a context window of 1,048,756 tokens, so you can send very large prompts in a single request, and a maximum output of 262,144 tokens. Its supported request parameters include temperature, top_p, top_k, min_p, frequency_penalty, presence_penalty, repetition_penalty, logit_bias, seed, stop, max_tokens, tools, tool_choice, reasoning, and include_reasoning. Because it exposes tools and tool_choice, it can be used in tool calling workflows, and the reasoning and include_reasoning parameters let you control whether reasoning output is returned.

Those capabilities point to a few practical uses. The provider describes LongCat 2.0 as suited for coding, repository-level changes, long-horizon problem solving, and agentic work. A large context window is useful when you need to pass whole files, long logs, or wide collections of source together, and the large maximum output helps when the answer itself has to be long, such as a substantial patch or a long report. Tool support makes it a reasonable candidate for agents that call functions or external services across several steps.

To call it through api14, use the OpenAI-compatible chat completions endpoint with the model id meituan/longcat-2.0, and include the parameters listed above in the request body where needed. Since api14 is OpenAI-compatible, existing OpenAI client libraries generally work with only the base URL and API key changed. Set max_tokens to bound the response, use stop sequences to end generation cleanly, and pass tools and tool_choice when you want the model to call functions. The seed parameter is available if you need more repeatable sampling, and temperature, top_p, top_k, and min_p give you control over how the model samples.

The main limits to keep in mind follow from the facts: input and output are text only, so this model is not the choice for image, audio, or other non-text input. If your task is short and simple, the very large context window and output limit may be more than you need, and a smaller text model could be a simpler fit. If you need multimodal input, look for a model whose input modalities include the media type you are sending.