Leaderboard / Humanizers / StealthGPT Super

September 2026 review · rank 1 of 10

StealthGPT Super AI detector test results

In the September 2026 HumanizerEval review, StealthGPT Super ranked 1 of 10 with a mean strict-bypass rate of 96.8%. It cleared all 5 detectors on 46 of 50 texts. Review our full results for the best AI humanizer on the leaderboard page.

StealthGPT Super results, September 2026

Website

stealthgpt.ai

Rank

1 of 10

Mean strict bypassThe average of the five detector pass rates: the share of texts each detector called human-written. The leaderboard is ranked by this number.

96.8%

Passed all 5 detectorsTexts that all five detectors called human-written.

46 of 50

Pangram 4.0 strict bypassHow many texts Pangram called human-written (a human score above 50%), with the mean and median human score it gave across all texts.

46 of 50 (92%), mean 88.9%, median 100%

GPTZero 2026-09-13-base strict bypassHow many texts GPTZero called human-written (a human score above 50%), with the mean and median human score it gave across all texts.

49 of 50 (98%), mean 99.0%, median 100%

Originality.ai turbo strict bypassHow many texts Originality.ai called human-written (a human score above 50%), with the mean and median human score it gave across all texts.

49 of 50 (98%), mean 95.8%, median 99.98%

ZeroGPT default strict bypassHow many texts ZeroGPT called human-written (a human score above 50%), with the mean and median human score it gave across all texts.

49 of 50 (98%), mean 91.1%, median 97.2%

Winston AI 5.0 strict bypassHow many texts Winston AI called human-written (a human score above 50%), with the mean and median human score it gave across all texts.

49 of 50 (98%), mean 97.7%, median 99.81%

Factual consistency passShare of outputs where key claims, numbers, citations, and named entities remain accurate and correctly attributed.

68%

Naturalness passShare of outputs where wording is fluent and free of awkward or clunky phrasing.

72%

Syntax integrity passShare of outputs where sentences are grammatically complete and free of disruptive run-ons or punctuation errors.

98%

Stance preservation passShare of outputs where the source's position, tone, and intent are not reversed or distorted.

96%

Mean length ratioOutput word count divided by input word count, averaged across texts. 1.00 is the same length; 1.10 is 10% longer.

1.04

Samples over 1.4× inputOutputs more than 1.4× the input’s word count.

0

Samples under 0.6× inputOutputs shorter than 0.6× the input’s word count.

0

Settings usedThe mode each tool ran on: its strongest advertised setting on the cheapest paid tier.

Super, production output

Recompute any of this from scores.csv. Strict bypass means an overall human verdict on a human score above 50%.

What the vendor claims, and what we measured

Claim

“StealthGPT passes Pangram AI detection at an 89% bypass rate based on internal testing conducted in September of 2026 and across hundreds of documents varying from academic to professional in writing style.” (stealthgpt.ai , retrieved September 18, 2026)

Measured

Strict bypass in September 2026: Pangram 92%, GPTZero 98%, Originality.ai 98%, ZeroGPT 98%, Winston AI 98%.

Verdict

Verified. Pangram strict bypass was 92% in September 2026, against the 89% claimed.

Review history

Head to head

Because every tool saw the same texts, these pages report per-sample agreement, how often each tool bypassed detection on the same input, rather than a difference of averages.

• StealthGPT vs WriteHuman

  StealthGPT cleared all 5 detectors on 14 samples where WriteHuman did not; the reverse happened on 1. Both cleared 31.

• StealthGPT vs Grubby

  StealthGPT cleared all 5 detectors on 29 samples where Grubby did not; the reverse happened on 1. Both cleared 17.

• StealthGPT vs Rephrasy

  StealthGPT cleared all 5 detectors on 26 samples where Rephrasy did not; the reverse never happened. Both cleared 20.

• StealthGPT vs Undetectable.ai

  StealthGPT cleared all 5 detectors on 31 samples where Undetectable.ai did not; the reverse never happened. Both cleared 15.

• StealthGPT vs RewriteAI

  StealthGPT cleared all 5 detectors on 33 samples where RewriteAI did not; the reverse never happened. Both cleared 13.

• StealthGPT vs HIX Bypass

  StealthGPT cleared all 5 detectors on 45 samples where HIX Bypass did not; the reverse never happened. Both cleared 1.

• StealthGPT vs Humbot

  StealthGPT cleared all 5 detectors on 46 samples where Humbot did not; the reverse never happened. Both cleared 0.

• StealthGPT vs BypassGPT

  StealthGPT cleared all 5 detectors on 46 samples where BypassGPT did not; the reverse never happened. Both cleared 0.

• StealthGPT vs Ryne

  StealthGPT cleared all 5 detectors on 46 samples where Ryne did not; the reverse never happened. Both cleared 0.