AI Capability Gap: Best U.S. Model vs. Best Chinese Model (ECI points)

## What this measures

How far the most capable AI model from a U.S. developer is ahead of the most capable model from a Chinese developer, in Epoch Capabilities Index points: ai.capability.eci_frontier.usa minus ai.capability.eci_frontier.chn for the same month. Each month's value is the highest ECI score among models released on or before the last day of that month (the current month reads to date). A model counts for a country when Epoch lists an organisation from that country among its developers. The series starts in January 2024: Epoch had scored one Chinese model by June 2023 and eleven by December 2023, so earlier months would show which models were scored rather than the frontier.

## Why it matters

It is the race in one number: whether China's best models are closing on the U.S. frontier or falling behind it.

## How to read it

Above zero, the best U.S. model scores higher. The gap jumps when either side releases a new best model, so read the trend over several months, not one step.

## What it does not say

It compares the single best model on each side, not the depth of each country's model lineup. It counts only models Epoch has scored, and a developer's country is where its organisation is based.

## Source

Palanor calculation from the Epoch Capabilities Index (Epoch AI, epoch.ai/eci, CC BY 4.0).