Grubby’s Academic mode label describes the setting used across the full mixed-genre sample. The two tools were tested on the same inputs. Grubby placed eighth on writing quality at 70.28, compared with Super’s third place at 73.48.
StealthGPT: Grubby AI Comparison and Alternative
Compare the tested humanizers on the same drafts, then choose the writing workflow you need.
StealthGPT Super scored higher overall and on writing quality than Grubby in Academic mode. Grubby passed all 50 Winston AI checks and Super passed 48. Super led on each of the other four detectors.
- 15 humanizers tested
- 50 matched inputs per tool
- Five detectors
HumanizerEval · November 2026
StealthGPT Super vs Grubby on 50 matched inputs
The November 17 run compared 15 humanizers on the same 50 English inputs. Detector rows report passes on each detector. Quality rows report provisional automated scores.
StealthGPT Super
Super, Quality, rephrase enabled, API
Grubby
Academic, API
| Measure | StealthGPT Super | Grubby |
|---|---|---|
| Overall score | 70.34/100 (#2) | 28.36/100 (#9) |
| Weighted detector pass rate | 69.00% | 10.40% |
| Writing quality | 73.48/100 (#3) | 70.28/100 (#8) |
| Pangram 4.1 | 44/50 (88%) | 1/50 (2%) |
| GPTZero 5.0b | 30/50 (60%) | 4/50 (8%) |
| Originality.ai 5.0 | 19/50 (38%) | 0/50 (0%) |
| Winston AI 5.2 | 48/50 (96%) | 50/50 (100%) |
| ZeroGPT | 34/50 (68%) | 22/50 (44%) |
| Readability | 68.80/100 | 57.20/100 |
| Meaning preservation | 67.20/100 | 67.60/100 |
| Factual consistency | 78.00/100 | 52.00/100 |
The overall score combines 70% weighted detector passes and 30% writing quality. Detector weights are 40% Pangram, 30% GPTZero, 20% Originality.ai, 5% Winston AI and 5% ZeroGPT. Winston uses a document Human Score of at least 80. The other detectors use qualifying human labels. HumanizerEval is operated by the StealthGPT team.
What the comparison shows
Frequently Asked Questions
Still have more questions?
Learn more in our support page.
The tested Grubby configuration used Academic mode through the API on the full 50-input set. It passed 1 Pangram check, 4 GPTZero checks and 50 Winston AI checks.
In November 2026, Grubby passed 1/50 (2%) on Pangram 4.1. StealthGPT Super passed 44/50 (88%) on the same 50 inputs. Only Pangram’s Human label counted as a pass.
HumanizerEval calculates the overall score using 70% weighted detector passes and 30% provisional automated writing quality. The quality rubric includes readability, factual consistency and six other measures. The scores apply to the configurations shown.
See how StealthGPT fits your writing workflow
Explore the AI Humanizer and see how Super revises a sample draft.

