Skip to content
RCreddit.com·

3.8-27B has ruined 3.5/3.6-35B’s for me. It’s just *absurdly* superior.

AI summary

A developer notes that the 3.8-27B model is "absurdly superior" to the 3.5/3.6-35B models, including vanilla, kat, Ornith/tiel, and nex-2 versions. The gap between 3.8-27B and these older models is significantly larger than the gap between 5.3 (flash and regular) and 3.8-27B. This superiority was observed across five replicated projects involving workflow design, data pipeline, results analysis, and data publishing.

Time & source

Times shown in UTC

Display time zone: UTC

Local time zone unavailable; showing UTC.

PublishedOffset at this time: UTC+0Sep 12, 2026, 10:16 UTC

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

Published
Sep 12, 2026, 10:16
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

Register to choose your default translation language in settings.

Applied science work, from workflow design, data pipeline, results analysis, article/reports writing and data publishing online. 5 projects I did in the past replicated from start to finish.

3x to 4x more total wall time. Yes, HUGE toll on how much you can do in a day if this was the only model you could use in your laptop.

But oh my…the quality of that thing. The stupid level of attention to detail. I have the Z.ai api, so I can compare it with 5.3 and 5.3-flash:

The gap between 5.3 (flash and regular) and 3.8-27B is much less, smaller when not plain tiny, than the gap between 3.8-27B and any of the 3.5/3.6-35B-A3B (vanilla, kat, Ornith/tiel, nex-2).

But it’s also spending 22 to 33% less tokens (effort =medium) and less ram footprint, so you get more done without hitting limits,compaction, etc.

So yeah, guess I’ll sip more tea, play the piano, whatever. Let that fat bottom Qwen work.

Source·reddit.com