# So what's the realy capable non-overthinking qwen 3.8 27b model? — retained record

> Retained reddit record from reddit:LocalLLaMA.

Canonical: https://highsignal.app/data/records/400df41b1ad775db

[← sources](https://highsignal.app/data)


source: [reddit](https://highsignal.app/data/reddit) (reddit:LocalLLaMA)  published 2026-09-10 18:08 ingested 2026-09-10 22:44

[original source ↗](https://www.reddit.com/r/LocalLLaMA/comments/1wcqwn1/so_whats_the_realy_capable_nonoverthinking_qwen/)

## retained text



```
Sorry folks, I am completely overwhelmed. There are just way too many variations of 3.8 27b available. The ones that I tried and more or less liked, are talking too much. The ones that are not thinking too much, are supposedly (?) not too capable. Is there some sort of concensus - like "**this** particular model is really good for coding/agentic, stable, **and** not too wordy"? Or should I just use vanilla unsloth + customized chat template?

PS Thanks everybody for replies, really appreciated! I got the point. It is **thinking** that makes qwen good. So I will just need to suck it up and learn to enjoy "wait let me reconsider" thing 😄 
```



document fetched 2026-09-11 00:44 · canonical https://reddit.com/r/LocalLLaMA/comments/1wcqwn1/so_whats_the_realy_capable_nonoverthinking_qwen
