Hugging Face Daily Digest Oct 25, 2026

The Hub is moving on quant work and preview models today. ISTA-DASLab puts a mixed precision GSQ-RCO build of DeepSeek-V4.1-Flash-1017 on the first default page. KRAFTON brings Raon-1-mini-preview to default trending with an official GGUF. Unsloth expands the MiniMax-M3.1-Flash GGUF ladder with five new Dynamic folders. Qwen OmniDR-Bench appears in Daily Papers.

On the Hub

ISTA-DASLab DeepSeek-V4.1-Flash-1017 GSQ-RCO GGUF

ISTA-DASLab/DeepSeek-V4.1-Flash-1017-GSQ-RCO-GGUF is new on default trending at #17. It is also #2 on GGUF trending. The card was created 2026-10-23T09:43:05Z. It is MIT licensed. The recommended IQ2_XXS build is 248.84 GB on disk. The authors run it in Strata on a single 12 GB GPU. The rest of the model lives in RAM. llama.cpp does not support DeepSeek-V4.1 as of Oct 23. ¹

KRAFTON Raon-1-mini-preview

KRAFTON/Raon-1-mini-preview is new on default trending at #26. It was created 2026-10-21T00:14:48Z. The model is Apache-2.0. It is a 30B-A4B hybrid MoE. It is a preview, not final. The official GGUF card offers Q4_K_M at 19.686 GB. ²

Unsloth MiniMax-M3.1-Flash GGUF

unsloth/MiniMax-M3.1-Flash-GGUF was last modified 2026-10-25T05:13:48Z. The tree grew from 22 files / 702.898 GB to 46 files / 1451.814 GB. Five new Dynamic folders were added. The smallest is UD-IQ1_M at 129.361 GB. The README is Work in progress. The license is minimax-community. ³

Daily Papers

Daily Papers shows WeVisOne-VL at 239 up. Qwen OmniDR-Bench is the named lab item. The paper is a video centered deep research benchmark. It has no Hub weights. ⁴