WILD CORPUS · reddit
PQS 29 (F) - prompt from www.reddit.com
Source: www.reddit.com · Scraped 2026-05-04 · Scored 2026-05-04
Score
F29 / 80
gemma4:latest · local · pqs-v2.0 · canonical
Clarity7 / 10
Specificity5 / 10
Context7 / 10
Constraints3 / 10
Output format3 / 10
Role definition1 / 10
Examples1 / 10
CoT structure2 / 10
The prompt
I am quite curious as I tried Gemma 4 31B, Qwen 3.6 27B, GLM 4.7 30B and some others in my native language (czech). Gemma performs "best" and considering the fact its "just" 18GB model - it actually blows my mind how well it can respond in my language. But lets say 1 in 50 words isnt correct. Very often its not even existing word, but its very similar to what i would expect to see. So its obvious that model tries to "remember" the correct word. So what about +- 100B models? How do they handle other languages than English and Chinese? As I am having quite a lot of fun and am not much restricted regarding money, i would like to know if getting more powerful hardware will bring the benefits. Thanks for responses - doesnt have to be about czech language, but some not so common like polish, magyar some yugoslavian languages ... whatever You tried.
This prompt was scraped from a public source. The score reflects the input as written, not the quality of any output it produced. The AI input quality problem is the gap between what people type and what the model can act on.