Key facts
- Meituan has open-sourced its LongCat-2.0 AI model, featuring 1.6 trillion parameters.
- The model was trained entirely on a cluster of 50,000 Chinese-made processors.
- LongCat-2.0 is designed for agentic coding, including code understanding, generation, and execution.
- It supports a native 1 million-token context window.
- Performance benchmarks indicate it is competitive with leading closed-source models like Gemini, GPT-5.5, and Claude Opus.
Chinese technology giant Meituan has released its LongCat-2.0 artificial intelligence model as open-source, a significant development as the system was trained entirely on domestic Chinese chips. The model boasts 1.6 trillion parameters and was developed using a cluster of 50,000 Chinese-made processors.
LongCat-2.0 is specifically designed for agentic coding tasks, including understanding, generating, and executing code. It features a native 1 million-token context window, allowing it to process vast amounts of information at once. The model utilizes a sparse mixture-of-experts (MoE) architecture, dynamically activating a subset of its parameters for efficiency during inference.
Meituan stated that the model's performance on benchmarks such as SWE-bench Pro and SWE-bench Multilingual is competitive with leading closed-source models, including Google's Gemini 3.1 Pro, OpenAI's GPT-5.5, and Anthropic's Claude Opus. Local semiconductor firms, including Huawei Technologies, Moore Threads, and MetaX, have confirmed their hardware supports the LongCat-2.0 system.