How far behind is China in AI? Measurements point to 7-8 months

How far ahead is the US, really, in the race for artificial intelligence? According to Epoch AI, an organization that analyzes AI development, Chinese models have trailed the American frontier by an average of seven months since 2023.

The figure is based on Epoch's own measurement tool, the Epoch Capabilities Index (ECI), which combines a model's abilities into a single score. The smallest lead in the analysis was four months, and the largest was 14 months.

The US has held the lead

According to Epoch, every model at the frontier of AI capabilities since 2023 has been developed in the US. As examples of top American models, Epoch cites GPT-5.2, Gemini 3 Pro and Claude Opus 4.5. Among the Chinese ones, it names DeepSeek-V3.2, Qwen3-235B-A22B-Thinking and Kimi K2 Thinking.

The first Chinese model to surpass GPT-4 on the ECI arrived in May 2024. That was 14 months after GPT-4 was released in March 2023. When the analysis was published in January 2026, no Chinese model had surpassed the ECI score of OpenAI's o3 from April 2025.

One explanation may lie in how the models are released. Almost all leading Chinese models are open-weight, meaning others can download and use the model itself. The leading American models are closed. The lead therefore roughly corresponds to the gap between closed and open models.

US test lands at eight months

In May 2026, CAISI, a unit of the US standards agency NIST, assessed that DeepSeek V4 Pro trails leading American models by about eight months. It is the most capable Chinese model CAISI has evaluated.

CAISI tested in cybersecurity, software development, math, natural sciences and abstract reasoning. DeepSeek V4 Pro came closer to the older GPT-5 than to Opus 4.6 and GPT-5.4. Math was the exception, where the model nearly matched the best American models.

The outlet The Decoder is critical of the evaluation and argues that CAISI likely has its own political agenda. According to The Decoder, measurements from Artificial Analysis show a fairly stable lead over time.

What does it mean?

The figures depend on who is measuring and how. Epoch's ECI and CAISI's evaluation use different methods and produce different leads. The gap between open and closed models is also not the same as the gap between China and the US. Overall, though, both measurements suggest that China is behind by somewhere between six months and a year.