# Is it just me or Kimi K3.1 Pro don’t look much better than DeepSeek 4.1 Flash? — retained record

> Retained reddit record from reddit:LocalLLaMA.

Canonical: https://highsignal.app/data/records/66cbc483062fb082

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


source: [reddit](https://highsignal.app/data/reddit) (reddit:LocalLLaMA)  published 2026-10-24 23:24 ingested 2026-10-25 02:39

[original source ↗](https://www.reddit.com/r/LocalLLaMA/comments/1xemxoq/is_it_just_me_or_kimi_k31_pro_dont_look_much/)

## retained text



```
Personally I am quite fond of the Kimi models, especially them  taking a gamble pushing the 2.8b parameter models. But that’s just it, you’d figure that the results would be representative of its size. However, the recent release of DeepSeek 1017 has put things in perspective for me. kimi k3 is a great model but realistically very hard to run locally, even when quantized, and even if managing to run it it is slow, due to the 104 active layers it need to go through to make a single token.  on the other hand, DeepSeek 4.1 flash 1017 results are quite close at a much smaller footprint and while also much faster. 

I have seen plenty coverage of the results of kimi k3 pro and to be frank, they don’t look like substantial improvement over DeepSeek 1017.

So thinking about it, maybe  just stick to 4.1 flash 1017 and ignore kimi k3 pro open weigh that will be released in nov 4th. Running it quantized will deteriorate its output  compared to a high quant version of DeepSeek. 

Just some thoughts I am having 
```



document fetched 2026-10-25 00:36 · canonical https://reddit.com/r/LocalLLaMA/comments/1xemxoq/is_it_just_me_or_kimi_k31_pro_dont_look_much
