China's Meituan released LongCat-2.0, a trillion-parameter LLM trained on 50,000 domestic chips with 1 million-token capacity, matching Gemini, GPT-5.5 and Claude Opus on coding benchmarks. Its agentic commerce features and chip self-sufficiency strategy intensify competitive pressure on DoorDash's AI-driven delivery business.
Meituan has released LongCat-2.0, its next-generation large language model with one trillion parameters and open-source access. The system supports up to one million token inputs, enabling processing of ultra-long documents and complex coding tasks.
LongCat-2.0 was trained from scratch on a cluster of 50,000 Chinese-made AI processors, marking a major step in reducing reliance on U.S. hardware. This strategy highlights China’s push for domestic chip self-sufficiency following export controls.
Benchmark tests show LongCat-2.0 matches or exceeds leading proprietary models—including Gemini, GPT-5.5 and Claude Opus—on coding and agent benchmarks. A preview version has already ranked among the top three models on a global AI marketplace.
Meituan’s agentic commerce capabilities—such as AI-powered ordering assistants and content generation—could heighten competition for DoorDash’s AI-driven delivery services. DoorDash may need to accelerate its own AI and infrastructure investments to maintain market position.
