ECI
A comprehensive evaluation benchmark introduced by Epoch AI that measures the aggregate capabilities of frontier large language models across complex reasoning, programming, mathematics, and science tasks. It is designed to reflect model performance on tasks approaching human expert-level difficulty.
Timeline (last 90 days)
Claude Opus 5.5, released by Anthropic, scored 167 on Epoch AI's ECI benchmark, ranking first in the world.
The cyber-ECI of some frontier models exceeds their general ECI by more than 10 points.
ECI (Effective Capability Index) is a method that converts model capabilities into a temporal progress measure, with its design philosophy drawing on the concept of purchasing power parity in economics.
ECI aims to address the problem of traditional benchmarks (such as MMLU, HumanEval, etc.) operating in silos and being difficult to compare horizontally, answering the question of what point in time a model's capabilities are equivalent to.
The typical pace of AI capability progress is approximately 15.5 ECI points per year, an empirical value derived from regression of the comprehensive performance of mainstream models across multiple benchmarks over the past several years.
All Facts (4)
Claude Opus 5.5, released by Anthropic, scored 167 on Epoch AI's ECI benchmark, ranking first in the world.
50%UnverifiedECI (Effective Capability Index) is a method that converts model capabilities into a temporal progress measure, with its design philosophy drawing on the concept of purchasing power parity in economics.
50%UnverifiedECI aims to address the problem of traditional benchmarks (such as MMLU, HumanEval, etc.) operating in silos and being difficult to compare horizontally, answering the question of what point in time a model's capabilities are equivalent to.
50%UnverifiedThe typical pace of AI capability progress is approximately 15.5 ECI points per year, an empirical value derived from regression of the comprehensive performance of mainstream models across multiple benchmarks over the past several years.
50%