NIST: DEEPSEEK V4 LAGS US MODELS BY 8 MONTHS

NIST's CAISI conducted a formal evaluation of DeepSeek V4 Pro, an open-weight AI model, in April 2026. The assessment determined the Chinese model operates at a performance level roughly equivalent to leading US AI systems from eight months prior. Despite the capability gap, DeepSeek V4 Pro represents a significant milestone for Chinese AI development. It surpasses all previous domestic models in performance metrics evaluated by NIST, establishing a new benchmark for the country's artificial intelligence sector. The evaluation provides a concrete timeline for comparing advanced AI systems across different regions. The eight-month lag reflects the rapid pace of AI development, where incremental improvements in training methods, data quality, and computational resources drive meaningful performance gains. NIST's CAISI, established to coordinate AI standards development across government and industry, regularly assesses frontier AI models to track progress and identify standardization needs. The DeepSeek V4 Pro evaluation contributes to this broader effort to create benchmarks for AI capability assessment. Open-weight models like DeepSeek V4 Pro differ from proprietary systems, allowing researchers and developers to inspect and modify the underlying code. This transparency has made the model popular among the AI research community, though the evaluation indicates performance still trails the most capable proprietary US systems in development. The findings underscore ongoing competition in large language models and generative AI between the US and China. Both nations continue investing heavily in AI research and infrastructure, with capability gaps narrowing over time as development techniques become more efficient and widely distributed.