What it means for leaders

Terminal coding agents spread developer by developer, often before any procurement decision. Check which ones your engineers already use, and set data and access rules before usage outgrows policy.

What would change the reading: Changes in the ranking between the three tools.

The trend

June 30, 2025September 30, 2026

The breakdown

September 30, 2026.
  • OpenAI Codex CLI89,664,560
  • Claude Code55,930,237
  • Gemini CLI1,629,189

Every reading

DateValueChangeSource
147,223,986−6.9%Source
158,111,827+39.5%Source
113,333,226+22.2%Source
92,711,572−66.5%Source
277,128,872+229.1%Source
84,200,700+35.4%Source
62,177,095+59.3%Source
39,038,978+22.3%Source
31,933,091+26.7%Source
25,210,200+2.2%Source
24,663,415−6.1%Source
26,260,242+0.9%Source
26,030,281+13.7%Source
22,893,120+27.2%Source
17,993,670+276.1%Source
4,784,246—Source

Definition and method

What it measures
Downloads in the month of @anthropic-ai/claude-code, @openai/codex and @google/gemini-cli from npm; parts give each tool.
Population
Downloads from the npm registry
Updated
Monthly
Source
npm registry download statistics · Public download statistics, attributed

Caveats

  • Downloads include automated installs and updates, so they are not users.
  • Tools installed other ways (native installers, package managers) are not counted.

Cite this

Enterprise AI in Numbers, "Monthly downloads of the major coding-agent command-line tools", https://lucianoayres.com/ai-brief/data/coding-agent-cli-downloads/, retrieved October 25, 2026. Data: npm registry download statistics.

Download every indicator (CSV) · Methodology