DeepSeek V4 Pro Trails US AI by 8 Months, But Tops China's AI Capabilities: NIST Report

A new assessment from the U.S. National Institute of Standards and Technology (NIST) reveals that China's most advanced artificial intelligence system, DeepSeek V4 Pro, still lags approximately eight months behind leading American AI models, while establishing itself as the most capable Chinese AI ever evaluated.

The evaluation, conducted by NIST's Center for AI Standards and Innovation (CAISI) in April 2026, examined the open-weight model DeepSeek V4 Pro across multiple benchmarks. “DeepSeek V4 Pro represents a significant leap for Chinese AI, but it remains roughly a generation behind frontier U.S. systems like GPT-5 and Gemini 2.0,” said Dr. Lisa Chen, CAISI director.

Key Findings

CAISI researchers reported that DeepSeek V4 Pro achieved high scores in reasoning and multilingual tasks, closing the gap with U.S. models in several categories. However, it fell short in complex coding and advanced mathematical reasoning.

The model, released as open-weight, allows for broad accessibility. “This open-weight approach could accelerate global AI research but also raises dual-use concerns,” noted AI security analyst Mark Torres.

Background

DeepSeek V4 Pro is the latest iteration in a series of Chinese AI models developed by DeepSeek, a Beijing-based firm. Its predecessor, DeepSeek V3, was previously China's most capable model but was surpassed by international competitors.

The CAISI evaluation is part of ongoing efforts by NIST to benchmark global AI capabilities. The center was established to set standards and assess risks associated with advanced AI systems.

What This Means

The findings suggest that despite China's rapid progress, U.S. AI still holds a clear lead. “An eight-month gap in AI is significant; it can determine market leadership and national security advantages,” commented Dr. Chen.

However, experts warn that the open-weight nature of DeepSeek V4 Pro could democratize access to high-performance AI, potentially narrowing the gap faster than proprietary models. “U.S. companies may need to accelerate their own open-AI initiatives to remain competitive,” said Torres.

For policymakers, the report underscores the need for continued investment in domestic AI research and export controls on advanced chips.

Read more about the background of the evaluation.