MEITUAN-W (03690) rose more than 4% in afternoon trading. As of the time of writing, the stock was up 4.23%, trading at HK$83.7, with a turnover of HK$2.045 billion. The movement follows industry news on April 24 that MEITUAN-W's new foundational large language model, LongCat-2.0-Preview, has entered open testing. The model boasts a total parameter scale exceeding one trillion, placing it among the world's top-tier large models. According to informed sources, the total parameter count and activated parameters of DeepSeek's newly released V4 model, launched the same day, are largely consistent with those of MEITUAN's LongCat-2.0-Preview. It is reported that both LongCat-2.0-Preview and DeepSeek V4 support a 1 million token context window, enabling the processing of millions of characters in a single inference—a capacity comparable to the newly released GPT-5.5. Furthermore, the new LongCat model has been deeply optimized for Agent application scenarios, effectively supporting production use cases such as code generation, complex task planning, and enterprise automation. Beyond its parameter scale, a more significant breakthrough of MEITUAN's new foundational model is that its entire training and inference process was completed using domestic computing power clusters. According to the same sources, the training phase utilized between 50,000 and 60,000 computing cards, making it the largest large model training task ever completed using domestic computing resources.
Comments