Meituan has released LongCat-2.0, an open source large language model containing 1.6 trillion parameters and supporting a 1 million-token context window. The model completed full process training on a computing cluster containing more than 50,000 domestic AI accelerators, making it the first trillion-parameter model to achieve that scale. LongCat-2.0 also completed the pre-training stage using domestic hardware, entirely avoiding the use of foreign AI hardware such as those from Nvidia. The system was built entirely on large AI ASIC superpods while using Huawei's Collective Communication Library. Meituan claims that LongCat-2.0 showed strong performance in coding and agent-based tasks while exceeding Google’s Gemini 3.1 Pro across several benchmarks, including Terminal-Bench 2.1 and SWE-Bench Pro. The company acknowledged that its latest model still trails OpenAI’s GPT-5.5 and Anthropic’s Claude 4.8 Opus on broader frontier capability assessments. Meituan faced major engineering difficulties, with memory becoming the primary bottleneck because each domestic accelerator offered substantially less capacity than Nvidia’s H800 chip.