November 2026: 15 services · five detectors · updated scoring

November 2026 · 750 outputs · 15 services

Pangram detector benchmark

Model: 4.1. Weight in the humanization composite: 40%.

Native Human classification only; Mixed and AI fail. Human score is the fraction of text classified human, not document confidence.

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.

Largest policy weight because independent research includes edited-AI detection evidence. This exact 40% is editorial, not a measured current accuracy. Reference. None of these weights is a measured current detector accuracy.

Pangram: ranked by strict-human pass rate, with raw averages separate.
HumanizerPasses/50Pass rateHuman-score average
Phrasly Ultra50/50100.00%100.00
StealthGPT Super44/5088.00%95.19
WriteHuman33/5066.00%76.92
Rephrasy10/5020.00%47.93
Undetectable.ai6/5012.00%36.35
ZeroGPT Humanizer3/506.00%21.10
Grubby1/502.00%8.68
Solvely1/502.00%7.09
Clever Humanizer0/500.00%6.48
BypassGPT0/500.00%0.73
Grammarly Pro0/500.00%0.00
Humbot0/500.00%0.71
RewriteAI0/500.00%6.30
QuillBot0/500.00%0.00
HIX Bypass0/500.00%0.00

Failed strict bypass

100 minus the pass rate on these humanized outputs; this is not a false-positive rate on human writing.

Pangram: failed strict bypass by service
HumanizerFailed/50Failed rate
Phrasly Ultra0/500.00%
StealthGPT Super6/5012.00%
WriteHuman17/5034.00%
Rephrasy40/5080.00%
Undetectable.ai44/5088.00%
Clever Humanizer50/50100.00%
ZeroGPT Humanizer47/5094.00%
BypassGPT50/50100.00%
Grubby49/5098.00%
Grammarly Pro50/50100.00%
Humbot50/50100.00%
Solvely49/5098.00%
RewriteAI50/50100.00%
QuillBot50/50100.00%
HIX Bypass50/50100.00%

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