# How smart is the IQ3 family of Qwen 3.8 Flash Next for coding tasks? — retained record

> Retained reddit record from reddit:LocalLLaMA.

Canonical: https://highsignal.app/data/records/e241533a294ac6e9

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


source: [reddit](https://highsignal.app/data/reddit) (reddit:LocalLLaMA)  published 2026-09-30 17:15 ingested 2026-10-01 02:44

[original source ↗](https://www.reddit.com/r/LocalLLaMA/comments/1wuark9/how_smart_is_the_iq3_family_of_qwen_38_flash_next/)

## retained text



```
I've been closely following the rise of the Strata inference engine and as someone with 28GB VRAM and 32GB RAM I'm itching to buy 32GB more RAM just to use Flash Next. But of course before I make such a financial commitment as a member of the GPU-poor class like myself, first I need to make sure the IQ3 quants are worth it. My use case is primarily agentic coding tasks with harnesses like Pi or OpenCode.

Has any of you guys used Flash Next IQ3 for relatively serious coding? Is it worth it? Compared to, say, Qwen 3.8 27B Q4 or Q5.

https://github.com/Niko1221/Strata/
```



document fetched 2026-10-01 00:45 · canonical https://reddit.com/r/LocalLLaMA/comments/1wuark9/how_smart_is_the_iq3_family_of_qwen_38_flash_next
