← sources

Qwen3.8-Flash-Next vs Swift-1.5-Qwen3.8-Flash-Next

source: reddit (reddit:LocalLLaMA)published 2026-10-26 02:35ingested 2026-10-27 02:40

retained text

Just switched over from DSV4 Flash 0731 to Qwen3.8-Flash-Next and I've been loving it. Its working significantly better to my surprise.

The only thing I can fault it with is unnecessarily deep work (thinking + fetching information) more often than not i.e. its not very token efficient so it feels slow. Swift1.5 is supposed to address this tendency of Qwen's models to overthink (likely RL'd that way to squeeze out maximum points on benchmarks). 

So looking for folk's experience before trying it myself. Anyone tried both? ran any anecdotal or better controlled comparisons?

document fetched 2026-10-27 00:43 · canonical https://reddit.com/r/LocalLLaMA/comments/1xfl3wo/qwen38flashnext_vs_swift15qwen38flashnext