LLM EXPLORER 58,608 MODELS INDEXED

Qwen3.8 27B MLX Mixed 4.95bpw by maglun

By maglun · 386 downloads
HF Repo ↗ Copy URL

Qwen3.8 27B MLX Mixed 4.95bpw is an open-source language model by maglun. Features: 27b LLM, VRAM: 16.6GB, License: apache-2.0, LLM Explorer Score: 0.32.

  4-bit   Apple-silicon Base model:quantized:qwen/qwen...   Base model:qwen/qwen3.8-27b   Conversational   Function-calling   Image-text-to-text   Mlx   Mlx-vlm   Quantized   Qwen3.8   Qwen3 5   Reasoning   Region:us   Safetensors   Sharded   Tensorflow   Thinking   Tool-calling   Vision
Model Card on HF 🤗: https://huggingface.co/maglun/Qwen3.8-27B-MLX-Mixed-4.95bpw 

Qwen3.8 27B MLX Mixed 4.95bpw Parameters and Internals

LLM NameQwen3.8 27B MLX Mixed 4.95bpw
Repository 🤗https://huggingface.co/maglun/Qwen3.8-27B-MLX-Mixed-4.95bpw 
Base Model(s)  Qwen/Qwen3.8-27B   Qwen/Qwen3.8-27B
Model Size27b
Required VRAM16.6 GB
Updated2026-08-16
Maintainermaglun
Model Typeqwen3_5
Model Files  5.4 GB: 1-of-4   5.3 GB: 2-of-4   5.3 GB: 3-of-4   0.6 GB: 4-of-4   0.9 GB
Model ArchitectureQwen3_5ForConditionalGeneration
Licenseapache-2.0
Model Max Length262144
Transformers Version5.8.0.dev0
Tokenizer ClassQwen2Tokenizer
Padding Token<|endoftext|>
Errorsreplace

Best Alternatives to Qwen3.8 27B MLX Mixed 4.95bpw

Best Alternatives
Context / RAM
Downloads
Likes
Qwen3.8 27B FP80K / 0.4 GB123157416
Qwen3.8 27B0K / 55.6 GB28753
Qwen3.6 27B0K / 55.4 GB67281382197
Qwen3.6 27B NVFP40K / 22 GB1887525419
...n3.8 27B MTPLX Optimized Speed0K / 19.6 GB698950
ThinkingCap Qwen3.6 27B0K / 54.7 GB35947614
Qwen3.6 27B FP80K / 0.4 GB7740472342
Qwen3.8 27B NVFP40K / 22 GB3012214
Qwen3.5 27B0K / 55.2 GB25362941028
Qwen3.8 27B INT8 W8A16 MTP0K / 30.7 GB472326
Note: green Score (e.g. "73.2") means that the model is better than maglun/Qwen3.8-27B-MLX-Mixed-4.95bpw.