I ran Qwen 3.8 Flash-Next on my AMD 7900 XTX at 500k context. All local and what a shift it has been.
I was running Qwen3.8-Flash-Next (base and later Swift-1.5 NVFP4) on a paid pod seat for about a month, and it started answering with short strings that echoed its own earlier reasoning.
Proof1 community thread
Key points
- I assumed the context was getting too big for it.
- Once I trusted Flash-Next again, I decided to check out the latest buzz of Flash-Next on Strata but had to see if it would work on AMD.
- An incredible improvement over my llama.cpp and Swift 1.5 27b dense.
- Set to 500k context in pi and now my daily driver.
Sources (1)
- [1]I ran Qwen 3.8 Flash-Next on my AMD 7900 XTX at 500k context. All local and what a shift it has been.r/LocalLLaMA (top, daily) · Oct 11, 09:44 PM
I was running Qwen3.8-Flash-Next (base and later Swift-1.5 NVFP4) on a paid pod seat for about a month, and it started answering with short strings that echoed its own earlier reasoning.
I assumed the context was getting too big for it.
Extractive summary: sentences quoted from the sources.
Before this
- Oct 11, 2026Taming the AI Hardware Cambrian Explosion — Abdul Dakkak, Modular
- Oct 11, 2026I tested different Qwen 3.8 27B quants
- Oct 11, 2026"Strata" for GLM5.3 Flash is here for some! Project Maya
- Oct 10, 2026Improve token per second without touching quant
- Oct 8, 2026Microsoft Joins the Local AI Push
- Oct 8, 2026ConwayResearch/Underdog-Saluki-27B-1.0
