Qwen3.8 27B GGUF & uncensored builds: size and RAM compared
Every popular Qwen3.8 27B repo on Hugging Face in one table: quantized GGUF and MLX files, full-precision weights and derived models, including abliterated and uncensored versions. GGUF builds load in llama.cpp, Ollama and LM Studio; MLX builds run on Apple-silicon Macs. Check the memory column to see what fits your machine.
Updated 2026-11-19 Β· 22 repos with 1,000+ downloads in 30 days Β· 16 GGUF Β· 1 MLX Β· 5 other Β· 8 uncensored Β· smallest needs ~17 GB
GGUF builds
MLX builds (Apple silicon)
| Model | Size | Quant | Memory | Uncensored | License | Downloads (30 days) |
|---|---|---|---|---|---|---|
| Qwen3.8-27B-MLX-4bit lmstudio-community | 27B | 4-bit | ~17 GB | β | Apache 2.0 | 3.6M |
Full-precision and other formats
| Model | Size | Quant | Memory | Uncensored | License | Downloads (30 days) |
|---|---|---|---|---|---|---|
| Qwen3.8-27B Qwen | 27B | 16-bit | ~65 GB | β | Apache 2.0 | 6.6M |
| Qwen3.8-27B-FP8 Qwen | 27B | FP8 | ~65 GB | β | Apache 2.0 | 6.3M |
| MiniMax-H3-Prompt-Rewriter-Qwen3.8-27B-Ref2VA lightx2v | 27B | 16-bit | ~65 GB | β | Not stated | 55.1K |
| Qwen3.8-27B-DSpine-b16 tencent | 27B | 16-bit | ~65 GB | β | Apache 2.0 | 6K |
| Qwen3.8-27B-pi-adapted harness-hub | 27B | 16-bit | ~65 GB | β | Apache 2.0 | 1.3K |
How this list is built
Hugging Face repos whose name contains βqwen3.8-27bβ, seen in the last 30 days with at least 1,000 downloads in the last 30 days, sorted by downloads. Size comes from Hugging Face metadata, or from the repo name when the metadata has none. Memory is an estimate: parameters Γ bytes per weight (about 0.5 for a 4-bit build, 2 for 16-bit) plus 20% headroom; long contexts need more. βGGUF (several)β means the repo offers several quant files.