Humanizer 12B GGUF vs. OptiQ 4-bit
How closely do two Humanizer 12B artifacts agree on recorded readability scores?
Across 33 matched rewrites with numeric readability scores, Humanizer 12B · GGUF scored higher on 16, Humanizer 12B · OptiQ 4-bit on 6, and 11 tied. Their November composite scores were 45.19 and 43.13 out of 100, respectively.
Results on the same source texts
Each row compares both entries on the same source and includes pairs with two numeric scores. Meaning uses recorded embedding cosine. Detector scores reflect reported human-likelihood.
| Metric | Humanizer 12B · GGUF higher | Humanizer 12B · OptiQ 4-bit higher | Tied | Measured pairs |
|---|---|---|---|---|
| Readability | 16 | 6 | 11 | 33 / 33 |
| Meaning preservation | 20 | 13 | 0 | 33 / 33 |
| GPTZero | 1 | 5 | 27 | 33 / 33 |
| ZeroGPT | 2 | 5 | 26 | 33 / 33 |
| Copyleaks | 0 | 6 | 27 | 33 / 33 |
| Originality.ai | 8 | 21 | 4 | 33 / 33 |
| Pangram | 14 | 13 | 6 | 33 / 33 |
| Grammarly | 11 | 13 | 9 | 33 / 33 |
| QuillBot | 4 | 19 | 10 | 33 / 33 |
Read the matched rewrites
These examples show the largest measured gaps in readability, with one example per writing category.
- Discussion board ? S21
Humanizer 12B · GGUF rewrite (80.0%) ? Humanizer 12B · OptiQ 4-bit rewrite (63.3%) - Marketing copy ? S31
Humanizer 12B · GGUF rewrite (81.3%) ? Humanizer 12B · OptiQ 4-bit rewrite (68.7%) - Academic essay ? S23
Humanizer 12B · GGUF rewrite (78.0%) ? Humanizer 12B · OptiQ 4-bit rewrite (68.7%)
Scope and source
Both entries use rewrites recorded from identified open-weight artifacts. The comparison uses the recorded outputs in the November 2026 dataset, version 2026-11-v1. See the methodology for metric definitions and limitations.
Model profiles: Humanizer 12B · GGUF and Humanizer 12B · OptiQ 4-bit.