DeepSeek V4 Pro lags frontier AI by eight months but offers cost-effectiveness

14 articles · Updated · NIST · May 2

CAISI’s April 2026 review said the open-weight model trailed the US frontier by about eight months across 16 benchmarks covering 35 models and five domains.
It found weaker results on some reasoning, software engineering and cyber tests, but said DeepSeek V4 was cheaper than GPT-5.4 mini on five of seven benchmark cost comparisons.
CAISI said DeepSeek’s own reported benchmarks made V4 appear closer to top US models, while its pre-committed suite suggested a broader US lead over China’s frontier.

If public benchmarks are flawed, how can we truly know which nation is winning the AI race?

Is China's cheaper AI a bigger long-term threat than its current eight-month capability gap?