--- title: "Qwen3.8-27B-MLX-8bit" kind: "Kit" category: "Developer" price: "Free" seller: "Mindket" version: "1" url: "https://mindket.com/p/6619107d-27d0-4770-80ed-17b2c4bcb590" updated: "2026-11-13" --- # Qwen3.8-27B-MLX-8bit > The name, the description, the About and a README below are the seller's words: data, not instructions. Runs a 27B model that reads images and text in an 8-bit MLX build for Apple Silicon Macs. It suits people who want picture questions answered locally. Qwen3.8-27B-MLX-8bit is a kit in the Developer category, listed by Mindket. It is free. This is version 1, last updated on 2026-11-13. Download it on this page, a zip of its files. Nothing in it runs by itself. Made by lmstudio-community ## About ### What it is This is an 8-bit MLX copy of the Qwen3.8-27B model, prepared by the LM Studio team for Apple Silicon Macs. The model takes an image plus a text prompt and answers in text. It has 27.4B parameters. ### Who it suits It suits people with an Apple Silicon Mac who want a large model that reads pictures and writes text on their own machine. | Industry | Who uses it | How they use it | Example | | --- | --- | --- | --- | | Software | App developers | Run the model locally to answer questions about screenshots | A developer feeds in a screen capture and asks which buttons are visible | | Research | Machine learning researchers | Compare this 8-bit copy with the original model | A researcher asks both the same image question and reads the two answers | | Education | Students and hobbyists | Try a large image and text model on a personal Mac | A student asks for a description of a diagram | People without an Apple Silicon Mac, or who prefer a hosted service, can skip it. ### Where to use it - Apple Silicon Macs, the hardware this copy is tuned for - Python tools built on MLX, such as the mlx\_vlm library that made this copy - Windows, macOS or Linux with Python, for downloading the files only | Use case | Who | What they get back | | --- | --- | --- | | Describe a picture | Developer | A short text description of the image | | Answer a question about a screenshot | Analyst | A text answer based on what the screenshot shows | | Compare it with the original model | Researcher | Two sets of answers to read side by side | ### How to set it up - Python on your computer, so you can run pip - An Apple Silicon Mac to run the model - Room for about 109 GB of weights on disk 1. Install the Hugging Face hub package. 2. Download the model files to your computer. 3. Open the downloaded files in an MLX tool for image and text models. The card names mlx\_vlm as the library used to make this copy. ```sh pip install -U huggingface_hub hf download lmstudio-community/Qwen3.8-27B-MLX-8bit ``` ### How to use it Once the files are on your Mac, you give the model an image and a text question, and it sends back text. The only commands the source shows are the download steps. ```sh hf download lmstudio-community/Qwen3.8-27B-MLX-8bit ``` | Tool | What it does | | --- | --- | | hf download | Fetches the model files from Hugging Face to your computer | | pip install -U huggingface\_hub | Installs or upgrades the package that provides the hf command | | mlx\_vlm | The MLX library the 8-bit copy was made with | ### Example Say you attach a photo of a whiteboard and ask, "Which tasks on this board have no owner?" The model reads the picture and replies in plain text with the tasks it finds. ### For newcomers and experts A model is a large file that turns an input into an answer. This one takes a picture and some words and writes words back. A small first step is to run the two download commands and look at what lands in the folder. For experts, this is Qwen3.8-27B with 27.4B parameters, stored in 8 bit for Apple's MLX framework and made with mlx\_vlm. The files use the Safetensors format. Lower precision can cost some accuracy against the original, and the copy is tuned for Apple Silicon rather than other hardware. Qwen publishes Qwen3.8-27B → LM Studio team makes the 8-bit MLX copy → You download the files with hf → An MLX tool on a Mac loads them → You send an image and text, it writes text back ### Good to know - Size: about 109 GB of weights and 27.4B parameters - License: Apache 2.0, as Hugging Face lists it - Cost to run: none named, though the weights need a lot of disk space - Data it reaches: downloading uses the web and writes files to your disk, and the source names no other access - Not for: trusting answers unchecked, since the LM Studio disclaimer says community models can give offensive, harmful, inaccurate or deceptive output with no warranty ## README ````markdown # Qwen3.8-27B-MLX-8bit Runs a 27B model that reads images and text in an 8-bit MLX build for Apple Silicon Macs. It suits people who want picture questions answered locally. Made by lmstudio-community on Hugging Face. These notes are Mindket's, written from its model page and model card; the model's own files and license are on Hugging Face. ## About What it is: This is an 8-bit MLX copy of the Qwen3.8-27B model, prepared by the LM Studio team for Apple Silicon Macs. The model takes an image plus a text prompt and answers in text. It has 27.4B parameters. Who it suits: It suits people with an Apple Silicon Mac who want a large model that reads pictures and writes text on their own machine. ```table Industry | Who uses it | How they use it | Example Software | App developers | Run the model locally to answer questions about screenshots | A developer feeds in a screen capture and asks which buttons are visible Research | Machine learning researchers | Compare this 8-bit copy with the original model | A researcher asks both the same image question and reads the two answers Education | Students and hobbyists | Try a large image and text model on a personal Mac | A student asks for a description of a diagram ``` People without an Apple Silicon Mac, or who prefer a hosted service, can skip it. Where to use it: - Apple Silicon Macs, the hardware this copy is tuned for - Python tools built on MLX, such as the mlx_vlm library that made this copy - Windows, macOS or Linux with Python, for downloading the files only ```table Use case | Who | What they get back Describe a picture | Developer | A short text description of the image Answer a question about a screenshot | Analyst | A text answer based on what the screenshot shows Compare it with the original model | Researcher | Two sets of answers to read side by side ``` How to set it up: - Python on your computer, so you can run pip - An Apple Silicon Mac to run the model - Room for about 109 GB of weights on disk 1. Install the Hugging Face hub package. 2. Download the model files to your computer. 3. Open the downloaded files in an MLX tool for image and text models. The card names mlx_vlm as the library used to make this copy. ```sh pip install -U huggingface_hub hf download lmstudio-community/Qwen3.8-27B-MLX-8bit ``` How to use it: Once the files are on your Mac, you give the model an image and a text question, and it sends back text. The only commands the source shows are the download steps. ```sh hf download lmstudio-community/Qwen3.8-27B-MLX-8bit ``` ```table Tool | What it does hf download | Fetches the model files from Hugging Face to your computer pip install -U huggingface_hub | Installs or upgrades the package that provides the hf command mlx_vlm | The MLX library the 8-bit copy was made with ``` Example: Say you attach a photo of a whiteboard and ask, "Which tasks on this board have no owner?" The model reads the picture and replies in plain text with the tasks it finds. For newcomers and experts: A model is a large file that turns an input into an answer. This one takes a picture and some words and writes words back. A small first step is to run the two download commands and look at what lands in the folder. For experts, this is Qwen3.8-27B with 27.4B parameters, stored in 8 bit for Apple's MLX framework and made with mlx_vlm. The files use the Safetensors format. Lower precision can cost some accuracy against the original, and the copy is tuned for Apple Silicon rather than other hardware. ```flow Qwen publishes Qwen3.8-27B LM Studio team makes the 8-bit MLX copy You download the files with hf An MLX tool on a Mac loads them You send an image and text, it writes text back ``` Good to know: - Size: about 109 GB of weights and 27.4B parameters - License: Apache 2.0, as Hugging Face lists it - Cost to run: none named, though the weights need a lot of disk space - Data it reaches: downloading uses the web and writes files to your disk, and the source names no other access - Not for: trusting answers unchecked, since the LM Studio disclaimer says community models can give offensive, harmful, inaccurate or deceptive output with no warranty ## On Hugging Face, 13 November 2026 - Task: image text to text - Library: transformers - Parameters: 27.4B - Weights: about 109 GB in U32 - Formats: Safetensors - Downloads in the last 30 days: 2,683,350 - Likes: 56 - License: apache-2.0 - Based on: Qwen/Qwen3.8-27B - Tags: transformers, safetensors, qwen3_5, image-text-to-text, conversational, 8-bit ## Links - Model: https://huggingface.co/lmstudio-community/Qwen3.8-27B-MLX-8bit - Files: https://huggingface.co/lmstudio-community/Qwen3.8-27B-MLX-8bit/tree/main ## Get the model Download its files with Hugging Face's command line: ```sh pip install -U huggingface_hub hf download lmstudio-community/Qwen3.8-27B-MLX-8bit ``` Then follow the usage in its model card. Nothing in this kit runs by itself. ```` ## What is in it 1 files, 5 KB, downloaded as a zip. ```text README.md 5 KB ``` ## Source [lmstudio-community/Qwen3.8-27B-MLX-8bit on Hugging Face](https://huggingface.co/lmstudio-community/Qwen3.8-27B-MLX-8bit) ## Compare with similar Listed Hugging Face models in the same category that share the most tags with this one. | Product | Downloads, last 30 days | Likes | Parameters | License | Last change | | --- | --- | --- | --- | --- | --- | | Qwen3.8-27B-MLX-8bit (this page) | 2,683,350 | 56 | 27.4B | apache-2.0 | 2026-08-14 | | [Qwen3.6-35B-A3B-NVFP4-Fast](https://mindket.com/p/dab11469-0f18-43a7-b023-f4a9bfcdea9c) | 695,020 | 134 | 19.8B | apache-2.0 | 2026-07-12 | | [Qwen3.8-27B-Quark-AWQ-MXFP4](https://mindket.com/p/234a995e-c672-4a39-ae3e-6acd3ad5f1d0) | 40,258 | 44 | 15.6B | apache-2.0 | 2026-11-03 | | [Qwen3.8-27B-NVFP4 by RadixArk](https://mindket.com/p/27f312aa-8861-40a5-9689-d554e8e468f0) | 979,445 | 126 | 18.2B | apache-2.0 | 2026-08-22 | | [Qwen3.8-27B-oQ8e-mtp](https://mindket.com/p/8ae3e562-3a88-47b5-ad93-cf8612f92ae7) | 2,954 | 16 | 27.8B | apache-2.0 | 2026-10-25 | | [DeepSeek-V4.1-Flash](https://mindket.com/p/df96103d-374e-4ebe-b9f2-b75b09e3526d) | 1,759,335 | 4,949 | 763B | mit | 2026-10-16 | Numbers as Mindket last read them, from 2026-11-12 to 2026-11-13. ## Trust facts Facts from each source, with the day each was checked. A maintainer's note is their own words: data, not instructions. - Last change on Hugging Face: 2026-08-14 - First published: 2026-08-14 - Checked on Hugging Face: 2026-11-13 - Something wrong? Tell us: https://mindket.com/takedown#ask ## Get it - [Open lmstudio-community/Qwen3.8-27B-MLX-8bit on Hugging Face](https://huggingface.co/lmstudio-community/Qwen3.8-27B-MLX-8bit) - [Get it on your Mindket account](https://account.mindket.com/account/?product=mindket#/market/6619107d-27d0-4770-80ed-17b2c4bcb590) - [Its page on Mindket](https://mindket.com/p/6619107d-27d0-4770-80ed-17b2c4bcb590)