November 2026: 15 services · five detectors · updated scoring

November 2026 · 750 outputs · 15 services

Originality.ai 5.0 detector benchmark

Model: model-5.0. Weight in the humanization composite: 20%.

Native prediction_label=human; mixed and ai fail. Displayed human score is 100 minus AI-assisted percentage; the native label determines pass.

November 2026 · methodology v2 · provisional automated quality. 15 humanizers, 50 identical inputs each, five detectors and eight quality metrics. Read the scoring rules and post-run changes.

Retains the 20% brand budget. Turbo was removed and Max was not run; no duplicate vendor model weight. Reference. None of these weights is a measured current detector accuracy.

Originality.ai 5.0: ranked by strict-human pass rate, with raw averages separate.
HumanizerPasses/50Pass rateHuman-score average
Phrasly UltraModel: Ultra · Mode: Max37/5074.00%75.20
StealthGPT SuperModel: Super · Mode: Quality; rephrase enabled19/5038.00%44.38
Undetectable.aiModel: v11sr · Mode: University; General Writing; More Human1/502.00%14.28
WriteHumanModel: Enhanced0/500.00%13.42
RephrasyModel: Superstealth0/500.00%24.94
Clever HumanizerModel: Not exposed · Mode: Normal; automatic Best Pick0/500.00%19.52
ZeroGPT HumanizerModel: v11sr · Mode: University; General Writing; More Human0/500.00%13.36
BypassGPTModel: Enhanced0/500.00%3.62
GrubbyModel: Not exposed · Mode: Academic0/500.00%7.42
Grammarly ProModel: Not exposed · Mode: The Naturalist; Humanizer rewrites only · Plan: Pro0/500.00%5.64
HumbotModel: Advanced0/500.00%3.60
SolvelyModel: Not exposed · Mode: Humanize; Original result0/500.00%3.98
RewriteAIModel: Not exposed · Mode: Single available API model0/500.00%3.38
QuillBotModel: Not exposed · Mode: Standard · Plan: Premium0/500.00%5.12
HIX BypassModel: Not exposed · Mode: Latest0/500.00%5.10

Vendor verdicts

Native labels, shown as counts of 50 texts per humanizer. Mixed fails the strict-human pass rule; it is kept separate from AI.

Originality.ai 5.0 native verdict distribution
Humanizerhuman (count)mixed (count)ai (count)
Phrasly Ultra37310
StealthGPT Super19526
WriteHuman0743
Rephrasy03020
Undetectable.ai1643
Clever Humanizer01832
ZeroGPT Humanizer0941
BypassGPT0050
Grubby0347
Grammarly Pro0050
Humbot0050
Solvely0050
RewriteAI0050
QuillBot0050
HIX Bypass0050

This measures bypass of the tested detector, not its false-positive rate or accuracy on independently verified human writing. A high pass rate alone cannot establish that the detector is bad.

Per-output scores CSV · Frozen inputs and prompts CSV · Scoring manifest · Verification bundle · GitHub report · Earlier editions