LongCat-2.0 outperformed Gemini 3.1 Pro on several coding benchmarks

LongCat-2.0 outperformed Gemini 3.1 Pro on several coding benchmarks

LongCat-2.0 contains 1.6 trillion parameters and a million-token contextMeituan trained the model using over 50,000 domestic AI acceleratorsThe model completed pre-training entirely without any Nvidia hardware involvedMeituan has released LongCat-2.0, an open source large language model containing 1.6 trillion parameters and supporting a 1 million-token context window.This scale places the model roughly on par with DeepSeek's flagship V4-pro, wh…

Behind the Delivery Riders: Meituan Quietly Built a Trillion-Parameter AI Model

Meituan, China leading local services platform best known for food delivery, has quietly achieved something remarkable: training a 1.6 trillion-parameter AI model called LongCat-2.0 running entirely on 50,000 domestic AI accelerator cards the world first trillion-parameter model trained and inferenced on China-made chips.The model scale places it among the largest language models globally. But more striking than the parameter count is what it re…

China’s Meituan open-sources AI model built on Chinese chips

Chinese tech giant Meituan has open-sourced a 1.6-trillion-parameter AI model trained entirely on domestic chips, reports Caixin. The move marks an industry first for a system of its size. The LongCat-2.0 model, released Monday for free commercial use, was developed using a computing cluster of 50,000 Chinese-made processors. Local semiconductor firms, including Huawei Technologies, Moore Threads, […]

Chinese DoorDash rival smashes open source record with 1.6-trillion-parameter LLM with a 1-million-context-token model crafted without Nvidia hardware

LongCat-2.0 contains 1.6 trillion parameters and a million-token context Meituan trained the model using over 50,000 domestic AI accelerators The model completed pre-training entirely without any Nvidia hardware involved 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. This scale places the model roughly on par with DeepSeek's flagship V4-pro…