November 2026: 16 services · five detectors · updated scoring

November 2026 · same 50 inputs

Rephrasy vs Superhumanizer

November 2026 · methodology v2 · provisional automated quality. 16 humanizers, 50 identical inputs each, five detectors and eight quality metrics. Read the scoring rules and post-run changes.

Updated November 18: Superhumanizer was added on the same 50 frozen inputs. The original 750 results were retained unchanged; no other humanizer was rerun.

November 2026: overall = 70% weighted bypass + 30% quality. All scores out of 100.
Overall rankHumanizerOverall averageOverall medianWeighted bypassQuality
4RephrasyModel: Superstealth37.4926.9527.60%60.561.11
15SuperhumanizerModel: Super Ultra (API engine) · Mode: Transform · Plan: Starter API19.4118.255.50%51.880.98

Rephrasy: detector breakdown

Strict passes on 50 outputs; raw human-score averages are shown separately and have different meanings across vendors.
DetectorModelWeightPassesPass rateNative human-score average
Pangram 4.14.140%10/5020.00%47.93
GPTZero 5.0b5.0b / 2026-11-04-base30%22/5044.00%38.24
Originality.ai 5.0model-5.020%0/500.00%24.94
Winston AI 5.25.25%44/5088.00%90.67
ZeroGPTVendor-managed; no public version identifier5%20/5040.00%66.05

Superhumanizer: detector breakdown

Strict passes on 50 outputs; raw human-score averages are shown separately and have different meanings across vendors.
DetectorModelWeightPassesPass rateNative human-score average
Pangram 4.14.140%1/502.00%4.62
GPTZero 5.0b5.0b / 2026-11-04-base30%2/504.00%5.57
Originality.ai 5.0model-5.020%0/500.00%4.96
Winston AI 5.25.25%31/5062.00%67.15
ZeroGPTVendor-managed; no public version identifier5%4/508.00%21.64

Overall weights humanization 70% and quality 30%. Rankings use automated quality scores with unresolved evidence flags; neither a detector pass nor a small score gap proves a universal advantage.

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