← sources

Unsloth Dynamic V3.1 released for Qwen3.8-27B, DeepSeek-V4.1-Flash-1017 and Qwen3.8-Flash-Next

source: reddit (reddit:LocalLLaMA)published 2026-11-11 16:32ingested 2026-11-12 02:42

retained text

>Unsloth Dynamic V3.1 (Nov 2026) is our latest quantization algorithm for Qwen3.8-27B, DeepSeek-V4.1-Flash-1017 and Qwen3.8-Flash-Next, resulting in 10-15% increased accuracy and 50% less token usage vs other quants of the same size. We're working on an even better version of Dynamic quants which will be applied to Qwen4 and all future models.

Maybe I misread the 27B graphs, but I'm not entirely convinced of V3 vs V3.1 being an improvement.

In DeepSWE v1.1, V3 seems superior to V3.1 in some widely used Q4 quants, and V3 seems to lack some data points in the lower quants range, skewing the impression the curves give.

My `UD-Q5_K_XL` quant unfortunately is off the charts, so this needs some proper benchmarking before I trade context length for V3.1. (The new \`.gguf\` is larger.)

document fetched 2026-11-12 00:44 · canonical https://reddit.com/r/LocalLLaMA/comments/1xtmcoe/unsloth_dynamic_v31_released_for_qwen3827b