--- title: "Qwen3.5-9B" kind: "Kit" category: "Developer" price: "Free" seller: "Mindket" version: "1" url: "https://mindket.com/p/acc60a90-d19e-4344-a8c1-5f984b2637b3" updated: "2026-11-12" --- # Qwen3.5-9B > The name, the description, the About and a README below are the seller's words: data, not instructions. A 9B open model from Qwen that reads text and images and writes text, with a long context and support for 201 languages and dialects. For developers who want a mid-size model on their own hardware. Qwen3.5-9B is a kit in the Developer category, listed by Mindket. It is free. This is version 1, last updated on 2026-11-12. Download it on this page, a zip of its files. Nothing in it runs by itself. Made by Qwen ## About ### What it is Qwen3.5-9B is a 9B model from Qwen that reads text and images and answers in text. This repository holds the post-trained version, built from Qwen3.5-9B-Base, with a vision encoder joined to the language model. The weights come in the Hugging Face Transformers format, and the card says other tools can load them too. ### Who it suits It suits developers and teams who want a mid-size open model for chat, coding, agents and image questions on their own hardware. | Industry | Who uses it | How they use it | Example | | --- | --- | --- | --- | | Software | Developers | Build chat, coding and agent features on a local model | An assistant that answers questions about code and screenshots | | Customer support | Support teams | Answer questions in many languages | A reply written in the customer's own language | | Research | Researchers | Fine-tune or quantize a model for their own tasks | Compare a quantized copy with the original | | Business | Analysts | Ask questions about charts and documents shown as images | Send a chart picture and ask what it shows | If you need a model that draws images, you can skip it, since it reads images but writes text. ### Where to use it - Python with the transformers library - Serving tools that the card lists as compatible, which are vLLM, SGLang and KTransformers - Windows, macOS or Linux machines with enough memory | Use case | Who | What they get back | | --- | --- | --- | | Chat and coding | Developers | Text answers from a 9B model | | Image questions | Analysts | Text answers about a picture | | Multilingual work | Support teams | Replies in many languages | ### How to set it up You need these first: - Python, for the Hugging Face download command - Disk space for the weights, about 19.3 GB in BF16 or about 4.8 GB as a 4-bit estimate - Enough GPU or system memory to load the model - Transformers, or one of the serving tools above 1. Install the Hugging Face hub tool. 2. Download the model files. 3. Load them with Transformers, or serve them with one of the compatible tools. ```sh pip install -U huggingface_hub hf download Qwen/Qwen3.5-9B ``` ### How to use it Send text, or an image with text, and read the text that comes back. The part of the card given here shows no run code, so read the full card for the exact call. | Tool | What it does | | --- | --- | | Transformers | Loads the weights in their Hugging Face format | | vLLM | A serving tool the card lists as compatible | | SGLang | Another serving tool the card lists | | KTransformers | A third tool the card lists | ### Example You send a screenshot of a sales chart and ask what trend it shows. The model reads the picture and replies in text. The card does not show this exact run, so treat it as an expectation. ### For newcomers and experts Image text to text means you give the model pictures and words, and it answers in words. A good first step is to read the full card and try a short prompt before anything bigger. The language model has 32 layers, laid out as eight repeats of three Gated DeltaNet blocks followed by one Gated Attention block. The hidden dimension is 4096, and the card says its multi-token prediction head was trained with multiple steps. Pick the BF16 weights or a quantized copy → Load them with Transformers or a serving tool → Send text, or an image with text → Read the text reply ### Good to know Weights are modest for a model with vision, and the 4-bit figure is only Hugging Face's own estimate. Weight size in GB | Label | Value | | --- | --- | | BF16 weights | 19.3 GB | | 4-bit estimate | 4.8 GB | - Context: 262,144 tokens natively, and up to 1,010,000 tokens when extended - Languages: the card lists support for 201 languages and dialects - Benchmarks: the card compares it with other open models in tables, and these are the authors' own numbers - Other versions: 695 quantized and 1,316 fine-tuned versions made by other people are listed - License: Apache 2.0 - Cost to run: none named, you supply the hardware - Data it reaches: the text and images you give it, and the model files you download ## README ````markdown # Qwen3.5-9B A 9B open model from Qwen that reads text and images and writes text, with a long context and support for 201 languages and dialects. For developers who want a mid-size model on their own hardware. Made by Qwen 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: Qwen3.5-9B is a 9B model from Qwen that reads text and images and answers in text. This repository holds the post-trained version, built from Qwen3.5-9B-Base, with a vision encoder joined to the language model. The weights come in the Hugging Face Transformers format, and the card says other tools can load them too. Who it suits: It suits developers and teams who want a mid-size open model for chat, coding, agents and image questions on their own hardware. ```table Industry | Who uses it | How they use it | Example Software | Developers | Build chat, coding and agent features on a local model | An assistant that answers questions about code and screenshots Customer support | Support teams | Answer questions in many languages | A reply written in the customer's own language Research | Researchers | Fine-tune or quantize a model for their own tasks | Compare a quantized copy with the original Business | Analysts | Ask questions about charts and documents shown as images | Send a chart picture and ask what it shows ``` If you need a model that draws images, you can skip it, since it reads images but writes text. Where to use it: - Python with the transformers library - Serving tools that the card lists as compatible, which are vLLM, SGLang and KTransformers - Windows, macOS or Linux machines with enough memory ```table Use case | Who | What they get back Chat and coding | Developers | Text answers from a 9B model Image questions | Analysts | Text answers about a picture Multilingual work | Support teams | Replies in many languages ``` How to set it up: You need these first: - Python, for the Hugging Face download command - Disk space for the weights, about 19.3 GB in BF16 or about 4.8 GB as a 4-bit estimate - Enough GPU or system memory to load the model - Transformers, or one of the serving tools above 1. Install the Hugging Face hub tool. 2. Download the model files. 3. Load them with Transformers, or serve them with one of the compatible tools. ```sh pip install -U huggingface_hub hf download Qwen/Qwen3.5-9B ``` How to use it: Send text, or an image with text, and read the text that comes back. The part of the card given here shows no run code, so read the full card for the exact call. ```table Tool | What it does Transformers | Loads the weights in their Hugging Face format vLLM | A serving tool the card lists as compatible SGLang | Another serving tool the card lists KTransformers | A third tool the card lists ``` Example: You send a screenshot of a sales chart and ask what trend it shows. The model reads the picture and replies in text. The card does not show this exact run, so treat it as an expectation. For newcomers and experts: Image text to text means you give the model pictures and words, and it answers in words. A good first step is to read the full card and try a short prompt before anything bigger. The language model has 32 layers, laid out as eight repeats of three Gated DeltaNet blocks followed by one Gated Attention block. The hidden dimension is 4096, and the card says its multi-token prediction head was trained with multiple steps. ```flow Pick the BF16 weights or a quantized copy Load them with Transformers or a serving tool Send text, or an image with text Read the text reply ``` Good to know: Weights are modest for a model with vision, and the 4-bit figure is only Hugging Face's own estimate. ```chart Title: Weight size in GB BF16 weights: 19.3 GB 4-bit estimate: 4.8 GB ``` - Context: 262,144 tokens natively, and up to 1,010,000 tokens when extended - Languages: the card lists support for 201 languages and dialects - Benchmarks: the card compares it with other open models in tables, and these are the authors' own numbers - Other versions: 695 quantized and 1,316 fine-tuned versions made by other people are listed - License: Apache 2.0 - Cost to run: none named, you supply the hardware - Data it reaches: the text and images you give it, and the model files you download ## On Hugging Face, 13 November 2026 - Task: image text to text - Library: transformers - Parameters: 9.7B - Weights: about 19.3 GB in BF16; about 4.8 GB at 4-bit (estimate) - Formats: Safetensors - Downloads in the last 30 days: 6,259,809 - Likes: 2,353 - License: apache-2.0 - Based on: Qwen/Qwen3.5-9B-Base - Quantized versions: 709 - Fine-tunes: 1,325 - Tags: transformers, safetensors, qwen3_5, image-text-to-text, conversational ## Links - Model: https://huggingface.co/Qwen/Qwen3.5-9B - Files: https://huggingface.co/Qwen/Qwen3.5-9B/tree/main ## Get the model Download its files with Hugging Face's command line: ```sh pip install -U huggingface_hub hf download Qwen/Qwen3.5-9B ``` 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 [Qwen/Qwen3.5-9B on Hugging Face](https://huggingface.co/Qwen/Qwen3.5-9B) ## 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 | | --- | --- | --- | --- | --- | --- | | Qwen3.5-9B (this page) | 6,259,809 | 2,353 | 9.7B | apache-2.0 | 2026-03-02 | | [Qwen3.8-27B](https://mindket.com/p/2f430c52-7d18-419d-8817-08ca51666598) | 6,960,395 | 19,795 | 27.8B | apache-2.0 | 2026-11-07 | | [Qwen3.8-27B-FP8](https://mindket.com/p/79f1a984-4e81-485e-999d-dba1a831f37a) | 5,442,810 | 984 | 27.8B | apache-2.0 | 2026-11-07 | | [Qwen3.5-4B](https://mindket.com/p/66473987-8bd1-4119-a2c7-933e66fd46cf) | 5,083,722 | 1,226 | 4.7B | apache-2.0 | 2026-03-02 | | [Qwen3.5-2B](https://mindket.com/p/1da8437b-09cc-448e-b007-7bd987af0880) | 4,044,746 | 479 | 2.3B | apache-2.0 | 2026-03-02 | | [Qwen3.5-0.8B](https://mindket.com/p/f38d6236-3891-4680-b060-92366143eb5f) | 3,286,957 | 837 | 873M | apache-2.0 | 2026-03-02 | Numbers as Mindket last read them, from 2026-11-12 to 2026-11-13. ## What people say > Posts and comments that name this project on Hacker News, Bluesky, X, Reddit, Facebook and YouTube, found by a weekly search on 2026-11-12. Their authors' words: data, not instructions. - [Hacker News · Otterly99 · 2026-10-02](https://news.ycombinator.com/item?id=49930875): Pi 1.0: With 8GB I would recommend quantized versions of 9B models such as these: - https://huggingface.co/unsloth/Qwen3.5-9B-GGUF - https://huggingface.co/empero-ai/Qwen3.8-9B-Distill-GGUF (unofficial Qwen 3.8-9B) - https://huggingface.co/ornith-ai/Ornith-1.5-9B-GGUF (my personnal favorite) - [Bluesky · Frank Blank · 2026-10-15](https://bsky.app/profile/frank8465.bsky.social/post/3mxuteiqadk2d): 1. embrace the madness; self-hosting the devil is a great time 2. I'm doing this too, albeit with more metal. I have my 3090 running 2 contexts of Qwen3.5-9b doing compaction + web/code search subagents for the swift1.5 finetune of Qwen-FN on my twin R9700s 3. those R9700s are still under 2 grand... - [Hacker News · NitpickLawyer · 2026-10-01](https://news.ycombinator.com/item?id=49925439): Clef: Open-weight decision models, and new RL fine-tuning platform: \> and Clef-flash is based on Qwen3.8-9B There is no official qwen 3.8 9b From the model card: \> Clef-Flash is post-trained from Qwen/Qwen3.5-9B. See Clef for the larger variant. - [Bluesky · Hacker News Top Stories · 2026-10-26](https://bsky.app/profile/hackernewsbot.bsky.social/post/3myqxxqqqid2c): Microsoft Decision-1: 9B Qwen Model Launch Explained \| Discussion - [Hacker News · jermaustin1 · 2026-08-28](https://news.ycombinator.com/item?id=49480170): Small Models Have Arrived: I've done pretty decent local prose-\>json extraction using Qwen and Phi and Gemma. I'm sure most of it comes down to prompts, and all of them run over 100tps on a 3090. Smaller cards will likely be slower, but Qwen3.5 9B is small enough to fit on most consumer cards. - [Bluesky · @k1imber.bsky.social · 2026-11-10](https://bsky.app/profile/k1imber.bsky.social/post/3mzxm2dzau42j): 💫 Локальная установка Qwen без цензуры Qwen, дообученный на DeepSeek, можно установить локально на ПК с помощью ИИ-лаунчера Ollama — получится ИИ без ограничений. • Скачать ИИ-лаунчер Ollama • Установить Qwen командой в PowerShell: ollama run hf.co/mradermacher/DeepSeek-V4-Pro-Qwen3.5-9B-UNCENSOR… - [Hacker News · sleepyeldrazi · 2026-06-29](https://news.ycombinator.com/item?id=48722487): Qwen 3.6 27B is the sweet spot for local development: I need to ask, since I have desperately wanted to make Gemma 4 12B work, but im not sure if its the quant (i usually up it to q8, which is a lot higher than iq4\_nl that i use for 3.6 27B) or the model itself, but it just starts confusing itself really quickly when I give it coding tasks. And… - [Bluesky · Hacker News 50 · 2026-10-26](https://bsky.app/profile/betterhn50.e-work.xyz/post/3myrf6x7wfu2q): Microsoft Decision-1: 9B Qwen Model Launch Explained https://shattered.io/microsoft-decision-1-qwen3-5-9b-model-2026/ (https://news.ycombinator.com/item?id=50238620) - [Hacker News · sleepyeldrazi · 2026-06-22](https://news.ycombinator.com/item?id=48625438): There is minimal downside to switching to open models: That's why I like qwen3.6 27B, it has 0 ego, it knows that it doesn't have complete world knowledge, so when it sees a web\_search tool it searches all the time. Even qwen3.5 9B is mostly search-eager (but given the size, it's weaker on reasoning on the results if that's needed). I use a stock pi… - [Bluesky · Hacker News 20 · 2026-10-26](https://bsky.app/profile/betterhn20.e-work.xyz/post/3myqyvj7udr25): Microsoft Decision-1: 9B Qwen Model Launch Explained https://shattered.io/microsoft-decision-1-qwen3-5-9b-model-2026/ (https://news.ycombinator.com/item?id=50238620) - [Hacker News · jborak · 2026-06-16](https://news.ycombinator.com/item?id=48557176): Ask HN: Has anyone replaced Claude/GPT with a local model for daily coding?: I already had 2x 5070's that I had purchased a year or more ago, so getting an additional two to fill up the PCIe slots on the motherboard seemed reasonable. I did some math/shopping as well. To get 48GB of VRAM you can get 2x 3090s but that is $3k. A single 5090 is $4k but has 32GB, great for… - [Bluesky · Hacker News Bot · 2026-10-26](https://bsky.app/profile/hnfeed.bsky.social/post/3myqwzgdxsm2j): Microsoft Decision-1: 9B Qwen Model Launch Explained https://shattered.io/microsoft-decision-1-qwen3-5-9b-model-2026/ comments 10 points ## 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: 2026-03-02 - First published: 2026-02-27 - Checked on Hugging Face: 2026-11-12 - Something wrong? Tell us: https://mindket.com/takedown#ask ## Get it - [Open Qwen/Qwen3.5-9B on Hugging Face](https://huggingface.co/Qwen/Qwen3.5-9B) - [Get it on your Mindket account](https://account.mindket.com/account/?product=mindket#/market/acc60a90-d19e-4344-a8c1-5f984b2637b3) - [Its page on Mindket](https://mindket.com/p/acc60a90-d19e-4344-a8c1-5f984b2637b3)