November 2026 · 750 outputs · 15 services
Winston AI detector benchmark
Model: 5.2. Weight in the humanization composite: 5%.
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.
Reduced 15% to 5% after this run at the benchmark owner’s request because most outputs passed. This is a disclosed post-run choice, not independent validation of detector weakness. Reference. None of these weights is a measured current detector accuracy.
| Humanizer | Passes/50 | Pass rate | Human-score average |
|---|---|---|---|
| Phrasly Ultra | 50/50 | 100.00% | 99.98 |
| WriteHuman | 50/50 | 100.00% | 99.93 |
| Undetectable.ai | 50/50 | 100.00% | 99.48 |
| Clever Humanizer | 50/50 | 100.00% | 99.84 |
| ZeroGPT Humanizer | 50/50 | 100.00% | 99.96 |
| BypassGPT | 50/50 | 100.00% | 99.41 |
| Grubby | 50/50 | 100.00% | 99.79 |
| Humbot | 50/50 | 100.00% | 99.49 |
| StealthGPT Super | 48/50 | 96.00% | 97.85 |
| RewriteAI | 48/50 | 96.00% | 97.41 |
| HIX Bypass | 46/50 | 92.00% | 95.48 |
| Rephrasy | 44/50 | 88.00% | 90.67 |
| Solvely | 44/50 | 88.00% | 86.95 |
| Grammarly Pro | 36/50 | 72.00% | 74.65 |
| QuillBot | 15/50 | 30.00% | 42.51 |
Failed strict bypass
100 minus the pass rate on these humanized outputs; this is not a false-positive rate on human writing.
| Humanizer | Failed/50 | Failed rate |
|---|---|---|
| Phrasly Ultra | 0/50 | 0.00% |
| StealthGPT Super | 2/50 | 4.00% |
| WriteHuman | 0/50 | 0.00% |
| Rephrasy | 6/50 | 12.00% |
| Undetectable.ai | 0/50 | 0.00% |
| Clever Humanizer | 0/50 | 0.00% |
| ZeroGPT Humanizer | 0/50 | 0.00% |
| BypassGPT | 0/50 | 0.00% |
| Grubby | 0/50 | 0.00% |
| Grammarly Pro | 14/50 | 28.00% |
| Humbot | 0/50 | 0.00% |
| Solvely | 6/50 | 12.00% |
| RewriteAI | 2/50 | 4.00% |
| QuillBot | 35/50 | 70.00% |
| HIX Bypass | 4/50 | 8.00% |
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