AI Capability Frontier (Epoch Capabilities Index)

## What this measures

The best general-capability score of any AI model released up to the end of each month, on the Epoch Capabilities Index (ECI). Epoch AI combines results from about 40 benchmarks (math, coding, science, reasoning, agentic tasks) into one scale, so models can be compared across years even after individual benchmarks top out.

## Why it matters

Capability is the first lever in every AI economic scenario: how much of today's work AI can do at all. This is the cleanest public read on how fast that frontier is moving.

## How to read it

Read the slope. A steeper climb means the frontier is advancing faster. The scale has no natural ceiling and no direct translation into "share of jobs"; compare the line with itself and with the open-weights frontier (ai.capability.eci_frontier_open). The current month reads "to date" and changes as models ship. Each point records which model held the frontier.

## What it does not say

Capability is not adoption. A higher score says what AI can do on benchmarks, not how much work it does in the economy. Benchmarks also miss things that matter at work: reliability over long tasks, judgment, physical tasks. Epoch scores models released from February 2023 on, so the series starts in March 2023 with GPT-4; earlier models such as GPT-3.5 are not scored. Epoch re-estimates past scores when it adds benchmarks, so the history is rewritten on each sync.

## Source

Epoch AI, Epoch Capabilities Index (epoch.ai/eci), from the benchmark_data.zip bundle at epoch.ai/benchmarks. Licensed CC BY 4.0: "Epoch AI's data is free to use, distribute, and reproduce provided the source and authors are credited." Palanor computes the monthly running maximum.