Write docs & copy
Pick the best model for docs and copy by human-like writing.
Right now GPT 5.6 Luna leads Human-like.
Sorted by compositeSynced 2026-10-10
| GPT 5.6 Luna | 100 | 1/1 | 1 |
| DeepSeek V4 | 97 | 1/1 | 2 |
| GPT 5.6 Sol | 92 | 1/1 | 3 |
| DeepSeek V4.1 Flash | 91 | 1/1 | 4 |
| GPT-6 Astra Pro | 89 | 1/1 | 7 |
| Gemini 3.1 Pro Preview | 89 | 1/1 | 6 |
| TInkling Small | 89 | 1/1 | 5 |
| Muse Spark 1.1 | 88 | 1/1 | 10 |
| Grok 4.5 | 88 | 1/1 | 9 |
| GPT 5.6 Terra | 88 | 1/1 | 8 |
| Qwen3.8 Max | 86 | 1/1 | 11 |
| HY4 Preview | 85 | 1/1 | 12 |
| GLM-5.3 | 82 | 1/1 | 14 |
| Qwen3.8 Flash | 82 | 1/1 | 13 |
| Qwen3.7 Flash | 81 | 1/1 | 17 |
| Grok 4.6 | 81 | 1/1 | 16 |
| DeepSeek V4 (0813) | 81 | 1/1 | 15 |
| Haiku 4.5 | 80 | 1/1 | 19 |
| Qwen3.7 Max | 80 | 1/1 | 18 |
| Kimi K2.6 | 77 | 1/1 | 20 |
Why it matters
Least slop is not the priciest
After inversion, GPT-5.6 Luna leads with a source Slop Score of 26. Pricier Fable 5 ($30/M) sits higher on the source board—more slop, not less.
People and machines both vote
40% is the live blind arena; 60% is mechanical distance from pre-2022 human writing: conciseness, templating, rhythm, and AI tells.
The same tasks in four domains
112 scenarios cover email, blog/essays, tweets, and Discord. A clean email writer is not always a clean essay writer.