AI detectors in the November benchmark
Pass rates and failed strict bypass rates on humanized text measure detector outcomes, not accuracy on verified human writing. See historical detector comparisons and their limitations.
Pangram
4.1 · 40% weight. Native Human classification only; Mixed and AI fail.
Human score is the fraction of text classified human, not document confidence.
GPTZero
5.0b / 2026-11-04-base · 30% weight. predicted_class=human; mixed and ai fail.
Human score is a class probability, not the percentage of human-written words.
Originality.ai 5.0
model-5.0 · 20% weight. Native prediction_label=human; mixed and ai fail.
Displayed human score is 100 minus AI-assisted percentage; the native label determines pass.
Winston AI
5.2 · 5% weight. Document human score≥80; explicit benchmark mapping, not a native categorical API verdict.
Winston returns a document Human Score. Sentence scores are not averaged to decide a pass.
ZeroGPT
Vendor-managed; no public version identifier · 5% weight. Unqualified Human Written or Most Likely Human written feedback passes. Mixed, uncertain and human labels qualified by possible AI fail.
Human score is 100 minus fakePercentage, a text-share estimate. It is not GPTZero.