Meituan open-sourced its trillion-parameter large-scale model LongCat-2.0, and simultaneously released the inference code for domestically developed Chinese card processors.
Related
MiniMax plans to launch a large-scale model with 2.7 trillion parameters.
Mars Finance reported on July 8th that Rare Earth Technology plans to launch a new generation of large-scale models with 2.7 trillion parameters. (Science and Technology Innovation Board Daily)
MiniMax plans to launch a large model with 2.7 trillion parameters.
PANews reported on July 8th that, according to sources, AI company MiniMax is developing a new large-scale language model with 2.7 trillion parameters, exceeding the scale of all currently available domestic AI models. The sources stated that this new model is expected to be released as early as the third quarter of this year. Internal employees involved in the development have designated it as M3 Pro, but it is unclear whether the company will use this name in the official release. MiniMax plans to open-source the model. This new model is significantly larger than MiniMax's current flagship model, M3 (428 billion parameters). Larger-scale AI models are better suited for handling complex reasoning and multi-step instruction tasks.
Meituan open-sources LongCat-2.0, simultaneously releasing inference code for domestically produced SIM cards.
Mars Finance reported on July 6th that Meituan officially open-sourced its trillion-parameter large-scale computing model, LongCat-2.0. LongCat-2.0 has a total of 1.6T parameters and an average activation of approximately 48B. It features deep collaborative optimization in model architecture, chip adaptation, and deployment strategies, specifically targeting domestically produced computing chips with limited GPU memory and bandwidth. (Wide Angle Observation)
Opinion: Open source models account for only 10% of enterprise large-scale model spending, but mature production environments will be dominated by open source models.
According to Beating's monitoring, while public opinion often touts that open-source large models are dominating everything, enterprise spending data presents the opposite picture. Jesse Zhang, co-founder and CEO of Decagon, an enterprise-level AI customer service platform, points out that the share of open-source models in total enterprise spending has now dropped to 11%. This decline stems from the fact that most enterprises' AI applications are still in the early, undefined exploratory stage, thus defaulting to reliance on closed-source models. However, he emphasizes that once application scenarios mature, open-source models will take over production environments with their advantages of extremely low latency and deep fine-tuning. In Decagon's own production environment, 90% of calls have already switched to open-source weighted models. The core driver of this transformation is interaction speed and customization capabilities, not cost savings. In customer service scenarios, a single conversation that takes 8 seconds to finish will completely destroy the product experience. Since leading closed-source labs do not allow fine-tuning of flagship models, and small closed-source models cannot be deeply customized, small-sized open-source models, through fine-tuning for specific tasks, have become the only option to support high-frequency real-time interactions. The future of enterprise AI will see a division of labor: leading closed-source labs will continue to dominate the exploration and discovery of new fields, while open-source weighted models will increasingly take over the actual production of mature businesses. Because model fine-tuning requires extremely high levels of data and talent, the migration from closed-source to open-source will be a slow process lasting several years, during which both will experience sustained growth.
Some large-scale AI models in China are 90% cheaper than those in the US; Chinese AI's high cost-effectiveness is capturing the US market.
According to a report by CNBC on July 7th, influenced by the continued price increases of models from leading US AI vendors, Chinese AI large-scale models are rapidly expanding their application scale in US enterprises due to their cost-effectiveness advantage. Industry insiders point out that the performance of some leading open-source and open weighted models in China is currently about 6 to 9 months behind the technology of top-tier US models such as OpenAI and Anthropic, while the price is 60% to 90% lower, and they can cover the vast majority of routine AI tasks, thus gaining popularity among US enterprises. According to statistics from the AI model aggregation platform OpenRouter, since February 8th of this year, the proportion of Chinese AI models used by US enterprises has exceeded 30% weekly, reaching a peak of 46%; while the average proportion in the previous 12 months was 11%. Another industry statistic shows that in the first week of the launch of Zhipu's latest large-scale model GLM 5.2, the daily average number of word calls increased by 27 times and the number of customers increased by 80 times, making it the fastest-deployed model on the platform in 2026; US AI startup Lindy has significantly reduced costs after switching all its AI business to DeepSeek models, and expects to save millions of dollars within a few months. (CCTV Finance)
Gu Yuxian, a Tsinghua University Special Scholar, joined DeepSeek, where he previously led the development of large-scale model distillation and a 50x speedup for long text processing.
According to Beating's monitoring, Gu Yuxian, a PhD graduate from the Department of Computer Science at Tsinghua University and recipient of the 2025 Graduate Special Scholarship, has officially joined DeepSeek, and his name has appeared in the author list of the DeepSeek V4 paper. Gu Yuxian's research mainly focuses on efficiency optimization of large models in the pre-training, model compression, and inference stages, and has been cited nearly 5,000 times on Google Scholar. Gu Yuxian's previous representative works include the knowledge distillation method MiniLLM for large models (which has been adopted by platforms such as Google, Alibaba, and NVIDIA), and the hybrid architecture model Jet-Nemotron. Jet-Nemotron achieves a 53.6 times faster throughput than traditional full-attention models when processing 256K ultra-long contexts on an H100 GPU, and surpasses hybrid expert models with larger parameter scales in multiple benchmark tests.