· 50 inputs · current edition
Rephrasy Review
Rephrasy ranks 4 of 15 on the 70/30 overall composite: 37.49/100. Weighted bypass is 27.60%; quality is 60.56/100.
What we tested
Model: Superstealth
Collection: API. The model name or selector is recorded; an additional provider build identifier was not exposed.
Superstealth; API. Every saved output was scored as delivered against its frozen source; no benchmark-side repair or detector-guided regeneration was applied. Settings describe this run, not a guarantee of the vendor’s strongest configuration for every use.
| Overall rank | Humanizer | Overall average | Overall median | Weighted bypass | Quality | |
|---|---|---|---|---|---|---|
| 4 | RephrasyModel: Superstealth | 37.49 | 26.95 | 27.60% | 60.56 | 1.11 |
AI detector results
| Detector | Model | Weight | Passes | Pass rate | Native human-score average |
|---|---|---|---|---|---|
| Pangram 4.1 | 4.1 | 40% | 10/50 | 20.00% | 47.93 |
| GPTZero 5.0b | 5.0b / 2026-11-04-base | 30% | 22/50 | 44.00% | 38.24 |
| Originality.ai 5.0 | model-5.0 | 20% | 0/50 | 0.00% | 24.94 |
| Winston AI 5.2 | 5.2 | 5% | 44/50 | 88.00% | 90.67 |
| ZeroGPT | Vendor-managed; no public version identifier | 5% | 20/50 | 40.00% | 66.05 |
Quality metrics
| Metric | Weight | Average |
|---|---|---|
| Readability | 25% | 52.40 |
| Meaning | 20% | 58.40 |
| Structure | 15% | 43.60 |
| Factual consistency | 10% | 64.00 |
| Completeness | 10% | 93.60 |
| Grammar syntax | 10% | 90.00 |
| Hallucinations | 5% | 2.00 |
| Stance preservation | 5% | 87.60 |
These automated scores are provisional. Evidence flags need review before treating individual model findings as confirmed. The headline quality score uses every dimension, including completeness and stance preservation.
Compare the result
Phrasly Ultra leads overall, Grammarly Pro leads quality, and StealthGPT Super is second overall. Use the individual detector results if one detector matters more to your workflow.
Compare humanizers · Full methodology
Frequently asked questions
How did Rephrasy perform in this benchmark?
Rephrasy ranks 4 of 15 on the 70/30 overall composite: 37.49/100. Weighted bypass is 27.60%; quality is 60.56/100.
Which Rephrasy settings were tested?
Model: Superstealth. Collection: API. Superstealth; API. We scored the saved output as delivered, without benchmark-side repair or detector-guided regeneration.
Did Rephrasy bypass GPTZero?
Rephrasy passed GPTZero on 22 of 50 frozen inputs (44.00%). Only the native human classification passes; mixed and AI fail. This is a result for this sample, not a guarantee.
Are these quality scores final?
49 of this service’s 50 outputs retain evidence-review flags. Scores are provisional automated assessments; flags are not confirmed defects and have not been human-adjudicated.
Can these scores be compared directly with older editions?
November uses five detectors, revised strict pass rules and eight weighted quality dimensions. Earlier editions used different panels and scoring. Compare within an edition; the earlier data remain archived.
50 anonymized humanizer results · Per-output scores CSV · Frozen inputs and prompts CSV · Scoring manifest · Verification bundle · GitHub report · Earlier editions