About Tool Benchmarks
Tool Benchmarks is an independent site that tests AI tools on fixed sets of texts and publishes every result, including the raw scores, so anyone can check our numbers or reuse them.
Tested
We started with AI text detectors because the question comes up every day in classrooms, newsrooms and content teams, and the answers people find are mostly vendor claims. A detector’s own marketing page will tell you its accuracy. It will rarely tell you how often it flags writing that a person wrote.
Our tests use writing whose origin is not in doubt. The human texts were all published before ChatGPT’s public launch on 30 November 2022, and the AI texts were written for this benchmark by models we name. Each text runs through every detector, and we publish the score each detector gave it.
The detectors tested so far are open-source models that anyone can download and run. We run them ourselves, on our own hardware, with the settings their authors publish. We do not accept payment from detector makers, and no company has seen our results before publication.
The methodology explains the test in full. The data page has every score for download.
Who publishes it
Tool Benchmarks is published by JMS Media Group LLC, a small independent company in Sheridan, Wyoming.