NIST CAISI Evaluates DeepSeek V4 Pro — Most Capable PRC Model, But 8 Months Behind Frontier

NIST·high signal

The Center for AI Standards and Innovation (CAISI) at NIST published its evaluation of DeepSeek V4 Pro, finding it the most capable Chinese AI model to date but approximately 8 months behind the frontier, performing similarly to GPT-5 (released ~8 months prior). CAISI found DeepSeek V4 more cost-efficient than the most competitive US model (GPT-5.4 mini) on 5 of 7 benchmarks, using non-public evaluations spanning cyber, software engineering, natural sciences, reasoning, and mathematics.