# By popular request: abliterated models — retained record

> Retained reddit record from reddit:LocalLLaMA.

Canonical: https://highsignal.app/data/records/6667912b95697f1f

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


source: [reddit](https://highsignal.app/data/reddit) (reddit:LocalLLaMA)  published 2026-11-18 20:54 ingested 2026-11-19 02:39

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

## retained text



```
Released a few more abliterated models by popular request.

These models have been around for quite a while and already have many abliterations so didn't know there was any demand for yet another abliteration. But multiple people have asked for it and folks [seem to prefer it over heretic](https://www.reddit.com/r/LocalLLaMA/comments/1xyv1oc/deepseek_v41_flash_1017_vs_mrfakename_abliterated/), etc.

So two new models now:

**GLM-5.3-Flash Abliterated**

[https://huggingface.co/mrfakename/GLM-5.3-Flash-abliterated](https://huggingface.co/mrfakename/GLM-5.3-Flash-abliterated)

[https://huggingface.co/mrfakename/GLM-5.3-Flash-abliterated-GGUF](https://huggingface.co/mrfakename/GLM-5.3-Flash-abliterated-GGUF)

**Qwen3.8-Flash-Next Abliterated**

[https://huggingface.co/mrfakename/Qwen3.8-Flash-Next-abliterated](https://huggingface.co/mrfakename/Qwen3.8-Flash-Next-abliterated)

[https://huggingface.co/mrfakename/Qwen3.8-Flash-Next-abliterated-GGUF](https://huggingface.co/mrfakename/Qwen3.8-Flash-Next-abliterated-GGUF)

  
The goal of my models is not to entirely remove *all* refusals. This can be done with aggressive abliteration, but nearly always leads to significant quality degradation on harmless prompts. 

Instead, with my abliterations I aim for a balance between usability and permissiveness. On nearly all prompts the model should not refuse; however, I did make the choice to accept a few refusals for a significantly more intelligent model.

For day-to-day use this shouldn't make a noticeable impact, the vast majority of refusals are still removed. I simply didn't optimize for getting to that "0/100" refusals number.

Of course, I understand that some may prefer for the model to retain absolutely zero refusals, in which case you are welcome to use the many other models available out there.

This is simply my personal preference :)

As always, feedback/thoughts/suggestions are greatly appreciated.
```



document fetched 2026-11-19 00:42 · canonical https://reddit.com/r/LocalLLaMA/comments/1xzqjlx/by_popular_request_abliterated_models
