November 2026 · 15 humanizers · 50 outputs each
Writing quality across AI humanizers
Grammarly Pro has the highest quality composite in this run at 83.38/100. This page ranks writing quality separately from detector bypass and the overall 70/30 score.
Quality leaderboard
Ranked by quality score. Metric definitions and weights
Scroll horizontally to see all eight metrics.
| # | Humanizer | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| 1 | Grammarly Pro | 83.38 | 65.20 | 79.60 | 90.40 | 90.00 | 99.60 | 100.00 | 78.00 | 94.80 |
| 2 | BypassGPT | 73.78 | 58.00 | 68.40 | 78.40 | 76.00 | 98.40 | 96.00 | 46.00 | 90.00 |
| 3 | StealthGPT Super | 73.48 | 68.80 | 67.20 | 62.00 | 78.00 | 94.00 | 100.00 | 32.00 | 94.80 |
| 4 | QuillBot | 72.46 | 63.20 | 75.60 | 81.20 | 66.00 | 98.80 | 70.00 | 26.00 | 91.60 |
| 5 | Solvely | 72.32 | 64.80 | 66.80 | 72.80 | 66.00 | 87.20 | 98.00 | 46.00 | 88.40 |
| 6 | Clever Humanizer | 71.94 | 59.60 | 69.20 | 77.60 | 70.00 | 97.20 | 72.00 | 60.00 | 92.80 |
| 7 | Phrasly Ultra | 70.80 | 59.20 | 64.00 | 68.40 | 76.00 | 99.20 | 94.00 | 24.00 | 96.40 |
| 8 | Grubby | 70.28 | 57.20 | 67.60 | 80.00 | 52.00 | 97.60 | 84.00 | 52.00 | 90.00 |
| 9 | Humbot | 68.78 | 56.80 | 65.20 | 75.60 | 48.00 | 98.40 | 94.00 | 36.00 | 87.20 |
| 10 | RewriteAI | 63.78 | 54.80 | 59.20 | 68.40 | 42.00 | 86.40 | 98.00 | 28.00 | 78.80 |
| 11 | Rephrasy | 60.56 | 52.40 | 58.40 | 43.60 | 64.00 | 93.60 | 90.00 | 2.00 | 87.60 |
| 12 | ZeroGPT Humanizer | 54.94 | 46.00 | 51.60 | 58.40 | 26.00 | 83.20 | 92.00 | 2.00 | 82.80 |
| 13 | Undetectable.ai | 53.76 | 45.60 | 50.40 | 52.40 | 26.00 | 82.00 | 96.00 | 0.00 | 80.40 |
| 14 | WriteHuman | 47.38 | 43.20 | 41.20 | 51.60 | 16.00 | 77.20 | 80.00 | 2.00 | 63.60 |
| 15 | HIX Bypass | 44.66 | 39.60 | 52.80 | 59.20 | 10.00 | 96.80 | 8.00 | 0.00 | 76.80 |
Score colors:0–2425–4950–6970–8990–100Higher scores are better; exact values are always shown.
What the eight metrics measure
Content fidelity accounts for 50% of quality, readability and grammar for 35%, and structure for 15%. Every metric below contributes to the quality composite; quality contributes 30% of the overall benchmark score.
| Metric | Weight | Definition |
|---|---|---|
| Readability | 25% | Assesses naturalness, clarity and fluent phrasing. Higher scores mean the rewrite reads more smoothly. |
| Meaning | 20% | Preserves the source’s main message, argument and key points. Higher scores mean the rewrite stays closer to the original meaning. |
| Structure | 15% | Assesses logical progression, paragraph organization and coherence between ideas. |
| Factual consistency | 10% | Keeps claims, names, numbers and relationships consistent with the source. This checks source consistency, not independent real-world fact verification. |
| Completeness | 10% | Retains the source’s substantive information, supporting details and important qualifications. |
| Grammar syntax | 10% | Assesses grammatical and syntactic quality, including sentence construction and correct language use. |
| Hallucinations | 5% | Avoids adding unsupported claims or details that are not in the source. Higher scores mean fewer unsupported additions. |
| Stance preservation | 5% | Keeps the original position, level of certainty and intent, without changing what the author is arguing. |
How to interpret these results
Scores are normalized to 0–100 and compare each delivered output against its frozen source. A score is an automated rubric assessment, not a probability that a text is correct. 602 outputs retain evidence-review flags; those flags are not confirmed defects, and scores remain provisional without independent human adjudication.
We publish definitions, weights and per-output grades so readers can verify the aggregate arithmetic. Proprietary evaluation instructions and implementation details are not published.
Quality methodology · Overall leaderboard
Per-output scores CSV · Frozen inputs and prompts CSV · Scoring manifest · Verification bundle · GitHub report · Earlier editions