Published humanizer benchmarks · 2026

Best AI humanizers of 2026: benchmark results

StealthGPT ranked #1 in HumanizerEval's October 2026 review and #2 overall with the highest readability score in HumanizerBench's November 2026 cycle. Compare the published results, test settings and detector-specific scores below.

HumanizerEval is built by the StealthGPT team and scores outputs through external detector APIs. Review its methodology and published per-sample results at humanizereval.com.

Highest readability in HumanizerBench's November 2026 cycle

StealthGPT's 77.7/100 readability score led all 16 tools in this cycle. It placed #2 overall at 69.47/100. The review evaluated 30 StealthGPT samples across 6 detectors.

Readability
77.7/100
#1 in the cycle
Overall placement
#2 of 16
69.47/100 overall score
StealthGPT samples
30
6 detector models

HumanizerBench is operated by WriteHuman. Readability measures clarity, fluency and naturalness through a language-model rating. The November cycle lists the Essential plan; its recorded model and mode fields are unspecified. The published methodology explains the overall score, detector results and quality measures.

HumanizerEval's October 2026 rankings

The HumanizerEval review ran 11 humanizers on the same 50 English texts. StealthGPT Super in quality mode ranked first by mean strict-bypass rate across 6 external detector models.

HumanizerMean bypassPangramGPTZeroOriginality.ai TurboOriginality.ai 5.0ZeroGPTWinston AI
1StealthGPT SuperMean bypass80.3%Pangram88%GPTZero92%Originality.ai Turbo92%Originality.ai 5.034%ZeroGPT84%Winston AI92%
2Undetectable.aiMean bypass71.3%Pangram40%GPTZero86%Originality.ai Turbo98%Originality.ai 5.04%ZeroGPT100%Winston AI100%
3WriteHumanMean bypass69.7%Pangram26%GPTZero90%Originality.ai Turbo100%Originality.ai 5.06%ZeroGPT96%Winston AI100%
4RephrasyMean bypass62.3%Pangram26%GPTZero70%Originality.ai Turbo86%Originality.ai 5.00%ZeroGPT92%Winston AI100%
5GrubbyMean bypass62.3%Pangram10%GPTZero68%Originality.ai Turbo100%Originality.ai 5.00%ZeroGPT96%Winston AI100%
6RewriteAIMean bypass56.3%Pangram2%GPTZero54%Originality.ai Turbo98%Originality.ai 5.00%ZeroGPT90%Winston AI94%
7PhraslyMean bypass48.7%Pangram4%GPTZero44%Originality.ai Turbo72%Originality.ai 5.00%ZeroGPT82%Winston AI90%
8HIX BypassMean bypass44.0%Pangram0%GPTZero6%Originality.ai Turbo66%Originality.ai 5.00%ZeroGPT92%Winston AI100%
9HumbotMean bypass44.0%Pangram0%GPTZero0%Originality.ai Turbo76%Originality.ai 5.00%ZeroGPT88%Winston AI100%
10BypassGPTMean bypass41.3%Pangram0%GPTZero4%Originality.ai Turbo56%Originality.ai 5.00%ZeroGPT88%Winston AI100%
11RyneMean bypass0.7%Pangram0%GPTZero0%Originality.ai Turbo0%Originality.ai 5.00%ZeroGPT4%Winston AI0%

Each detector column is the share of 50 texts that the detector gave a human score above 50%. The best rate on each detector is in bold. With 50 texts per tool, gaps of a few points are within the margin of error. How the test works

Compare the detector results

StealthGPT Super ranked #1 on the mean. The per-detector results help you choose which tests to include in your own evaluation:

  • Originality.ai Turbo: 4 tools scored higher, up to 100%, against 92% for StealthGPT.
  • ZeroGPT: 8 tools scored higher, up to 100%, against 84% for StealthGPT.
  • Winston AI: 8 tools scored higher, up to 100%, against 92% for StealthGPT.

It passed all 6 detectors on 17 of 50 texts. Its lowest rate was on Originality.ai 5.0, which 33 of its 50 texts did not pass.

Detectors disagree

On the same texts, the average tool passed Winston AI 89% of the time and Originality.ai 5.0 4% of the time, so a tool's rate on one detector can be far from its mean.

See which detectors were hardest to pass