Huihui Qwen3.8 27B abliterated GGUF is a proof-of-concept, uncensored GGUF conversion of Qwen/Qwen3.8-27B created by abliteration, distributed by huihui-ai under apache-2.0. The repository offers several GGUF series derived from different base GGUF sources, including bf16, UD, UD-DW, GSQ-RCO, and Ternary variants. Abliteration is applied to selected layer ranges, with differences between updates: the latest update ablates layers 17 to 52 (0-based indexing), while update 4 and later update 5/6 series reportedly ablate only layers 22 to 52. The base model card lists pipeline_tag image-text-to-text and notes that MTP and visual components were not modified. The ternary hybrid-attention files require the PrismML-Eng/llama.cpp fork and will not run with stock llama.cpp. Some weights below Q8_0 were converted to Q8_0 (filenames K_L), Q8_0 weights were changed to BF16 (Q8_0_L), and the resulting file sizes may differ from the original GGUF. The model card labels this as a test/validation and a crude proof-of-concept implementation.
Format
Q4_K_M GGUF weights
Freshness 79 out of 100. Repository age: 60 days old, Aug 16, 2026. Measures age, not quality.Freshness79/10060 days old
Popularity 85 out of 100. 2,329,638 downloads in 30 days; 1,330 lifetime likes. Checked Oct 15, 2026. Measures adoption, not quality.Popularity85/1002.3M downloads / 30d
Qwen3.8-27B GGUF is a GGUF-quantized version of Qwen/Qwen3.8-27B, produced with ByteShape's ShapeLearn method. The published model card lists an Apache-2.0 license, image-text-to-text pipeline, the gguf tag, and base model Qwen/Qwen3.8-27B. ShapeLearn is described as learning per-tensor datatypes intended to maintain quality at low bitlengths, and this release is said to replace earlier ShapeLearn-Lite quantizations. The card states models went through a multi-benchmark evaluation suite covering math, coding, general knowledge, instruction following, and agentic tool use in thinking and instruct modes against a BF16 baseline. It also describes vision capability, a bundled MTP head, llama.cpp usage, speculative decoding options, a build requirement note for DFlash 2, GPU-optimized quantizations, quality-vs-TPS charts on RTX 5090 with links to additional plots, and a model table beginning with GPU-1 IQ2_XXS at 2.5 bits/weight.
Format
IQ4 GGUF weights
Freshness 80 out of 100. Repository age: 58 days old, Aug 18, 2026. Measures age, not quality.Freshness80/10058 days old
Popularity 80 out of 100. 1,556,746 downloads in 30 days; 173 lifetime likes. Checked Oct 15, 2026. Measures adoption, not quality.Popularity80/1001.6M downloads / 30d
Arena Open Source Text: rank 23 of 225. Ranked model: qwen3.8-27b. Original model ranking, not a benchmark of this package. Export dated Oct 13, 2026. Source checked Oct 15, 2026. Human preference ranking, not hardware suitability.#23Arena · Open SourceTextOriginal model ranking+3
A dense vision-language model for coding, reasoning, and agent tasks, with image and video understanding.
Model size
27B
Format
Ollama 27B package
Context
262,144 native
Freshness 76 out of 100. Repository age: 71 days old, Aug 5, 2026. Measures age, not quality.Freshness76/10071 days old
Popularity 95 out of 100. 7,363,968 downloads in 30 days; 17,749 lifetime likes. Checked Oct 15, 2026. Measures adoption, not quality.Popularity95/1007.4M downloads / 30d
Arena Open Source Web coding: rank 10 of 55. Ranked model: qwen3.8-27b. Export dated Oct 13, 2026. Source checked Oct 15, 2026. Human preference ranking, not hardware suitability.#10Arena · Open SourceWeb coding+3
An experimental open-weight multimodal model with sparse attention and 262K native context.
Model size
125B language · 6B active
Context
262,144 native
Freshness 82 out of 100. Repository age: 52 days old, Aug 24, 2026. Measures age, not quality.Freshness82/10052 days old
Popularity 88 out of 100. 2,290,500 downloads in 30 days; 6,235 lifetime likes. Checked Oct 15, 2026. Measures adoption, not quality.Popularity88/1002.3M downloads / 30d
Arena Open Source Web coding: rank 3 of 55. Ranked model: qwen3.8-flash-next. Export dated Oct 13, 2026. Source checked Oct 15, 2026. Human preference ranking, not hardware suitability.#3Arena · Open SourceWeb coding+1