# Is anyone actually using deepseekv4.1f at sub 1-3 bpw? — retained record

> Retained reddit record from reddit:LocalLLaMA.

Canonical: https://highsignal.app/data/records/df22d76b640af705

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


source: [reddit](https://highsignal.app/data/reddit) (reddit:LocalLLaMA)  published 2026-10-28 12:59 ingested 2026-10-29 02:42

[original source ↗](https://www.reddit.com/r/LocalLLaMA/comments/1xhndoi/is_anyone_actually_using_deepseekv41f_at_sub_13/)

## retained text



```
Need a quick sanity check since I been seeing a lot of posts about engines that run deepseekv4.1 1017 at low bpws on affordable hardware but with surprisingly usable speeds.

Is anyone actually using them for anything at all beyond treating it like a novelty? The last time I tried deepseek at sub q4 was v4-0731's Q2, and that thing was very much a shadow of it's full weight Q8 self. And the Q8 was basically 90% QAT-ed Q4 anyways, and it did not take well to going down 2 more bpw.

If you are using a 1-3 bpw v4.1-1017 can you share your experience with it? Preferably in contrast to another model like 3.8FN at higher quants. 
```



document fetched 2026-10-29 00:44 · canonical https://reddit.com/r/LocalLLaMA/comments/1xhndoi/is_anyone_actually_using_deepseekv41f_at_sub_13
