China Approves 'Fenghe V1.0', World's First 100-Billion Parameter Weather AI Model, for Operational Use
On September 9, the China Meteorological Administration (CMA) announced that 'Fenghe V1.0', a 100-billion-parameter domain-specific large language model for weather services, passed expert review and achieved official operational deployment approval. Developed by CMA alongside Xiong'an AI Innovation Research Institute and Zhipu AI, the model specializes in weather service understanding, content generation, meteorological reasoning, and tool invocation. As the world's first 100-billion-parameter weather service LLM to enter operational deployment, Fenghe demonstrates superior performance over general-purpose LLMs in domain reasoning and risk assessment. Its international version is integrated into China's 'MAZU' smart early warning framework, supporting the UN Early Warnings for All initiative to enhance global climate resilience. Fenghe V1.0 was trained on 50 million tokens of high-quality meteorological service corpus combined with authoritative weather datasets, and it has completed registration with the Cyberspace Administration of China. Additionally, the developers have open-sourced the full model weights globally on platforms including GitHub, Hugging Face, and ModelScope.
## BACKGROUND
Traditional weather forecasting relies heavily on numerical weather prediction models that solve physics equations using high-performance supercomputing. Domain-specific AI models supplement these systems by rapidly processing multi-source observational data, interpreting complex weather patterns, and generating tailored disaster warning advisories for the public and emergency responders.