Study: Best non-US, non-China AI models trail US by nearly 10 months: what are the key facts?

Research institute Village compared 857 models worldwide using a unified capability score to measure each country's gap with the US.

Study: Best non-US, non-China AI models trail US by nearly 10 months: what are the key facts?

Since 2024, the strongest models outside the US and China have reached equivalent performance an average of 9.9 months later than the US.

Study: Best non-US, non-China AI models trail US by nearly 10 months: what are the key facts?

Over the same period, new Chinese models lagged the US by an average of 5.2 months, about half the gap of other countries.

Study: Best non-US, non-China AI models trail US by nearly 10 months: what are the key facts?

In 2026, labs outside the US and China placed only about 20 models on the list, versus 113 for the US and 79 for China.

Study: Best non-US, non-China AI models trail US by nearly 10 months: what are the key facts?

South Korea and Mistral lag by about 10 months, while India's Sarvam trails by about 28 months, the largest gap.

Physix Frontier · News Briefing Card (enbrief · Nov 7, 2026)

Study: Best non-US, non-China AI models trail US by nearly 10 months

KEY FACTS

Research institute Village compared 857 models worldwide using a unified capability score to measure each country's gap with the US.
Since 2024, the strongest models outside the US and China have reached equivalent performance an average of 9.9 months later than the US.
Over the same period, new Chinese models lagged the US by an average of 5.2 months, about half the gap of other countries.
In 2026, labs outside the US and China placed only about 20 models on the list, versus 113 for the US and 79 for China.
South Korea and Mistral lag by about 10 months, while India's Sarvam trails by about 28 months, the largest gap.

KEY DATA

9.9 monthsMedian lag of non-US, non-China models behind the US

5.2 monthsMedian lag of Chinese models behind the US

113/79/20Models on the 2026 list (US/China/other)

about 8 monthsMistral Large 4 gap

PHYSIX OBSERVATION

Sovereign AI has been talked up for years, but the reality is that most countries are standing still: neither left behind nor catching up. The ones actually narrowing the gap, South Korea and Mistral, are still grinding away at general-purpose large models, while those pivoting to platforms or data centers fall behind. This suggests that beyond compute and capital, the organizational ability to keep training frontier models is the real dividing line, and for small countries, betting on the application layer may be the more pragmatic path.

Source: enbrief original report ↗