Realtime AI News
Meituan Releases LongCat-2.0 Trillion-Parameter Model, Joining China's Mega-Model Race
Meituan has officially launched LongCat-2.0, a trillion-parameter large language model, marking the company's entry into the ultra-large-scale AI model competition. The release signals accelerated strategic investment in AI infrastructure and in-house model development by the food delivery and local services giant.
According to a weekly technology roundup published by the Beijing Municipal Science and Technology Commission, Meituan has released LongCat-2.0, a trillion-parameter LLM. This represents a major upgrade from the original LongCat model, pushing the parameter count into the trillion-parameter league.
Meituan has not yet disclosed specific technical details of LongCat-2.0, including architecture design, training dataset size, or benchmark performance. However, the trillion-parameter scale places it in the top tier of Chinese large models, directly competing with Baidu's ERNIE, Alibaba's Tongyi Qianwen, and ByteDance's Doubao.
Training a trillion-parameter model demands extraordinary compute, data, and engineering resources. Meituan's advantage lies in its vast multimodal data from local services — food delivery, dining, travel, and retail — which could give LongCat-2.0 a differentiated edge in vertical applications.
The strategic significance of this release is that it marks the full participation of all major Chinese internet companies in proprietary AI model development. Unlike general-purpose conversational models, Meituan's model has a natural fit for service-commerce and lifestyle scenarios, potentially delivering advantages in recommendation systems, customer service, and dispatch optimization.
However, the trillion-parameter trend also raises questions about practicality — bigger models don't always translate linearly to better business outcomes. Finding effective integration points between ultra-large models and real business scenarios will be a core challenge for Meituan.
Key developments to watch include whether LongCat-2.0 will be offered via public API or deployed primarily for internal business use, and whether consumer-facing AI assistants for local services will emerge from this model foundation.
Why it matters
Meituan's entry into the trillion-parameter model race means virtually all major Chinese internet companies now have proprietary large models, shifting competition from parameter counts toward real-world scenario deployment.
Nearby Updates
All07/06, 10:45
Beijing No. 1 Token Factory Phase 1 Begins Production in Yizhuang, Marking a New Era for AI Infrastructure
Phase 1 of the Beijing No. 1 Token Factory has commenced production in the Yizhuang district, marking a significant milestone in China's AI infrastructure development. The facility is designed to provide efficient, low-cost token processing capabilities for large-scale model training and inference.
07/06, 10:54
China's LLM Weekly Token Usage Hits 23.45 Trillion, DeepSeek-V4-Flash Leads for Seventh Consecutive Week
China's large language model market recorded 23.45 trillion tokens in weekly usage, maintaining high momentum for the tenth straight week. DeepSeek-V4-Flash topped the rankings for the seventh consecutive week, while Zhipu AI's GLM-5.2 broke into the top five. Chinese model vendors swept the top six spots.
07/06, 09:45
WAIC 2026 Preview: Post-Scaling Paradigm Shift as Chinese AI Agents March Toward Production Deployment
WAIC 2026, scheduled for July 17-20 in Shanghai, will focus on the post-scaling era paradigm restructuring, with AI agents as the core driver of industrial deployment. The conference will showcase China's indigenous tech stack including MemTensor memory architecture and the Harness engineering framework, marking a shift from model competition to agent productivity.
07/06, 12:06
OpenAI and Anthropic Speed Toward IPOs as Token Payment Model Faces Customer Backlash
Anthropic, valued at $965 billion in May, is slated to debut as early as October, with OpenAI likely to follow in 2027. Palantir CEO Alex Karp publicly criticized the token payment model, while Uber, Microsoft, Salesforce, and Meta have started rationing employee AI usage amid soaring costs.