U.S. AI risk management agency CAISI reports that DeepSeek V4 Pro lags U.S. leading AI models by about 8 months but is currently the highest-performing Chinese AI model

NIST’s CAISI report says DeepSeek V4 Pro trails the latest major U.S. AI models by about eight months overall across 5 areas and 9 benchmarks.

The model was still rated around GPT-5 level, making it the strongest Chinese model in the assessment.

On cost efficiency, DeepSeek V4 Pro beat OpenAI’s GPT-5.4 mini on several benchmarks.

The report highlights both the U.S.-China AI capability gap and how government evaluators weigh performance versus cost.