Us Vs China Eci

This January 2026 data insight from Epoch AI puts a number on the US-China race in AI capability (epoch.ai). Its finding: since 2023, every model at the top of Epoch's capability rankings came from the United States, and Chinese models trailed them by an average of seven months. The gap was never smaller than four months or larger than 14.

The yardstick is the Epoch Capabilities Index, or ECI, a single score that Epoch builds by combining a model's results across many benchmarks. To turn scores into a time gap, Epoch takes the best Chinese model on each day and asks how long ago the leading US model was at the same level or below, treating scores within one ECI point as equal. It drops the earliest model in each country's series (LLaMA-65B and Baichuan1-7B) because its data starts in January 2023 and those models were probably not at the true frontier. The 14-month high point came in May 2024, when a Chinese model first passed GPT-4. At the time of writing, no Chinese model had surpassed the ECI score of OpenAI's o3, released in April 2025.

Epoch notes that this gap closely resembles the broader gap between closed and open-weight models, which it measured separately in its open-weight comparison. It calls that resemblance unsurprising, since nearly all leading Chinese models release their weights openly while frontier US models stay closed.