Skip to content
RCreddit.com·

Qwen-Next seems worse to me then 3.8 27b for coding, but I feel like I must be missing something?

AI summary

A user running Qwen-Next and 3.8 27b models on MTPLX with 128GB RAM for coding tasks, specifically in pi agent in nodejs, finds that 3.8 27b performs better on harder tasks. Despite reading that Qwen-Next is superior, the user's experience with the q8 27b model, possibly running as a dynamic q4 with 8-bit attention due to "optimized for speed" settings, suggests otherwise, leading them to question if they are misinterpreting something.

Time & source

Times shown in UTC

Display time zone: UTC

Local time zone unavailable; showing UTC.

PublishedOffset at this time: UTC+0Sep 12, 2026, 00:05 UTC

IngestedOffset at this time: UTC+0Sep 12, 2026, 15:01 UTC

Published
Sep 12, 2026, 00:05
Ingested
Sep 12, 2026, 15:01
Source type
Dev community
Tier
Community
Source status
Healthy

Tier is a per-source editorial setting, not a per-item score.

Article

Hi! I run both models on MTPLX on my m5 max, and since I have 128GB of ram I run the q8 27b. I think MTPLX only lets me run "optimized for speed" which it says is a dynamic q4 with 8 bit attention.

Both of them honestly are very speedy! For coding (in pi agent in nodejs) I've just noticed that 27B feels stronger with harder tasks. But I've read so many people on here say qwen-next is better so I was wondering if maybe I'm just doing or thinking about it wrong?

(and p.s. its sooo amazing that alibaba just made and released this amazing models for free! ❤️)

Source·reddit.com