AI Research Velocity (arXiv)

## What this measures

New papers submitted to arXiv in the month and listed in Artificial Intelligence (cs.AI), Machine Learning (cs.LG) or Computation and Language (cs.CL), each paper once, by first-version submission. Complete months only.

## Why it matters

Research output precedes capability, capital spending and the pricing pressure that follows; paper counts are the earliest public read on the pace of the AI research frontier.

## How to read it

The series starts in January 2010 at 60. The latest reading, for September 2026, is 11,262. The high is 12,583 (May 2026) and the low 46 (January 2011).

## What it does not say

Counts papers, not their quality or influence. Conference deadlines make some months jump; compare a month with the same month a year earlier.

## Source

arXiv API counts of papers by category and first-version submission date, complete months only; arXiv monthly submission statistics. Metadata CC0 1.0. Thank you to arXiv for use of its open access interoperability. Source: arXiv.