# List of suggestions for Qwen4 models — retained record

> Retained reddit record from reddit:LocalLLaMA.

Canonical: https://highsignal.app/data/records/b07a92eb28660492

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


source: [reddit](https://highsignal.app/data/reddit) (reddit:LocalLLaMA)  published 2026-10-29 10:45 ingested 2026-10-30 02:40

[original source ↗](https://www.reddit.com/r/LocalLLaMA/comments/1xigkze/list_of_suggestions_for_qwen4_models/)

## retained text



```
There is a list of suggestions for Gemma 5 [post](https://www.reddit.com/r/LocalLLaMA/comments/1xhtg61/list_of_suggestions_for_gemma_5/). So let me kick a similar discussion for Qwen4 and most probably Qwen4.X models. I know that it may lead to nowhere, but if there is a chance to be heard by someone from Qwen team and see some of the suggestions implemented in Qwen4.1 models why not to try. Here is my list and I will update the post with most voted features from the comments.

1. Context size 512K or more.
2. Less reasoning + caveman style in reasoning.
3. Small and medium sized models.
4. Make a model answer in simple plain English.

**Most voted suggestions form the comments:**

1. OSS license apache/mit

I think the first three points are obvious and I would like to elaborate on the fourth. I have really hard time reading pages of highly technical stuff with all the details when I ask such simple questions, like "How can I run X model on this engine faster" and want to get simple actionable answer. Instead I get 4 pages of technical details, which will take me more time to process than 30 mins of manual googling. It is as if you bring your car to service station and after the question what is wrong with my car you get 30 mins technical lecture like: “Your vehicle’s lead-acid electrochemical storage system exhibits advanced sulfation of the electrode surfaces, reducing the active area available for charge transfer and substantially increasing internal resistance. Under the starter motor’s high-current load, this produces a precipitous voltage drop, preventing the delivery of sufficient torque to initiate crankshaft rotation...” When the answer should be “Your battery is worn out, so the car needs a new one to start.”

Why am I writing all these, because I feel much more cognitively overloaded now with AI then in pre-AI era. You may argue that in coding it is good to get more detailed information. But why do you need it if nowadays almost no one writes code manually and let's be honest may be less than 10% of users even look at generated code. So it is complication for complication. If you really need a more detailed answer just write "Elaborate your answer on this and this" and you will get what you want. And yes, I tried to change model style with prompting and by installing style extension in pi. In first case it works 1-2 turns and reverts back to its usual style. In second QFN in UD-Q4\_K\_XL breaks almost immediately after few tool calls and starts to loop indefinitely.

Edt: Added suggestion from the comments.
```



document fetched 2026-10-30 00:45 · canonical https://reddit.com/r/LocalLLaMA/comments/1xigkze/list_of_suggestions_for_qwen4_models
