--- title: "NorthCUA-Mini" kind: "Kit" category: "Developer" price: "Free" seller: "Mindket" version: "1" url: "https://mindket.com/p/50dfec70-c66f-4693-9f0a-887c6469974b" updated: "2026-11-12" --- # NorthCUA-Mini > The name, the description, the About and a README below are the seller's words: data, not instructions. Reads screenshots and requests, then plans and issues actions to operate a desktop or browser. A 35B mixture-of-experts computer-use model with 3B active parameters, built on Qwen3.5. NorthCUA-Mini 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 beipo-ai ## About ### What it is NorthCUA-Mini is a computer-use model from beipo-ai, built on Qwen3.5. You feed it a request along with screenshots and past steps. It answers with its reasoning, a reply and tool calls aimed at a desktop or a browser. It has 35B total parameters, of which 3B are active, and sits in the middle of the NorthCUA family. ### Who it suits Developers and researchers building agents that operate software through the screen, in desktop apps or the web. | Industry | Who uses it | How they use it | Example | | --- | --- | --- | --- | | Software | Agent developers | Drive a desktop or browser through screenshots and actions | An agent that fills in a web form | | Quality assurance | Test engineers | Script user flows without writing selectors | A run through a checkout page | | Research | AI researchers | Study computer-use agents trained on long task trajectories | A comparison across the Nano, Mini and Pro sizes | | Operations | Business automation teams | Automate repeat steps in desktop applications | Moving data between two programs | If you want a plain chat model, or have no runtime to execute actions, you can skip it. ### Where to use it - Desktop applications, driven through screenshots - Web workflows in a browser - Windows, macOS or Linux with Python, for downloading the files | Use case | Who | What they get back | | --- | --- | --- | | Desktop tasks | Agent builders | Reasoning and tool calls for the next action | | Web tasks | Automation teams | Steps that move through browser pages | | Benchmark study | Researchers | A mid-size model to compare with Nano and Pro | ### How to set it up - Python, to run the Hugging Face download tool - About 70.2 GB of disk for the BF16 weights, or about 17.6 GB at 4-bit as an estimate - The official NorthCUA companion runtime the card names, which the model needs for computer use 1. Install the Hugging Face hub tools with pip. 2. Download the model files with the hf command. 3. Set up the official companion runtime the card links. 4. Connect the model to it for prompts, tool calls, observations and actions. ```bash pip install -U huggingface_hub hf download beipo-ai/NorthCUA-Mini ``` ### How to use it You run the model together with its official companion runtime, which sends screenshots and history in and carries the actions out. All GUI coordinates use a 0 to 1000 normalized scale instead of raw pixels, and the runtime converts them at execution time. ### Example You ask the agent to open a settings page in a browser and switch on dark mode. The model looks at the screenshot, reasons about where the control is, and returns a tool call with a coordinate on the 0 to 1000 scale. The runtime clicks it. ### For newcomers and experts Start by reading the card and getting the companion runtime working on a harmless test task. Do not skip it, because the card says the model and runtime are designed to be used together. The card says the family learned from long trajectories with dependent steps and outcome checks. Training used staged learning, with grounding on SFT checkpoints and task completion on later joint desktop and web RL checkpoints. The 3B active figure reflects work done per token, so loading the full weights still takes far more memory. User request and screenshot go in → Model reasons about the screen → It returns a tool call, 0 to 1000 coordinates → Runtime converts to pixels and acts → A new screenshot comes back ### Good to know - License: Apache 2.0 - Cost to run: none named, though the weights are about 70.2 GB in BF16 - Data it reaches: screenshots and interaction history of the desktop or browser it controls, plus Hugging Face for the download - Not for: use without the official companion runtime, which the card calls required for computer use - Limits: the card says its reported results come from the technical report and are not new measurements of these files NorthCUA-Mini reported scores in percent | Label | Value | | --- | --- | | ScreenSpot-Pro | 92.41% | | OSWorld-G | 87.94% | | OSWorld-Verified | 85.30% | | WebArena-Verified | 70.43% | ## README ````markdown # NorthCUA-Mini Reads screenshots and requests, then plans and issues actions to operate a desktop or browser. A 35B mixture-of-experts computer-use model with 3B active parameters, built on Qwen3.5. Made by beipo-ai 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: NorthCUA-Mini is a computer-use model from beipo-ai, built on Qwen3.5. You feed it a request along with screenshots and past steps. It answers with its reasoning, a reply and tool calls aimed at a desktop or a browser. It has 35B total parameters, of which 3B are active, and sits in the middle of the NorthCUA family. Who it suits: Developers and researchers building agents that operate software through the screen, in desktop apps or the web. ```table Industry | Who uses it | How they use it | Example Software | Agent developers | Drive a desktop or browser through screenshots and actions | An agent that fills in a web form Quality assurance | Test engineers | Script user flows without writing selectors | A run through a checkout page Research | AI researchers | Study computer-use agents trained on long task trajectories | A comparison across the Nano, Mini and Pro sizes Operations | Business automation teams | Automate repeat steps in desktop applications | Moving data between two programs ``` If you want a plain chat model, or have no runtime to execute actions, you can skip it. Where to use it: - Desktop applications, driven through screenshots - Web workflows in a browser - Windows, macOS or Linux with Python, for downloading the files ```table Use case | Who | What they get back Desktop tasks | Agent builders | Reasoning and tool calls for the next action Web tasks | Automation teams | Steps that move through browser pages Benchmark study | Researchers | A mid-size model to compare with Nano and Pro ``` How to set it up: - Python, to run the Hugging Face download tool - About 70.2 GB of disk for the BF16 weights, or about 17.6 GB at 4-bit as an estimate - The official NorthCUA companion runtime the card names, which the model needs for computer use 1. Install the Hugging Face hub tools with pip. 2. Download the model files with the hf command. 3. Set up the official companion runtime the card links. 4. Connect the model to it for prompts, tool calls, observations and actions. ```bash pip install -U huggingface_hub hf download beipo-ai/NorthCUA-Mini ``` How to use it: You run the model together with its official companion runtime, which sends screenshots and history in and carries the actions out. All GUI coordinates use a 0 to 1000 normalized scale instead of raw pixels, and the runtime converts them at execution time. Example: You ask the agent to open a settings page in a browser and switch on dark mode. The model looks at the screenshot, reasons about where the control is, and returns a tool call with a coordinate on the 0 to 1000 scale. The runtime clicks it. For newcomers and experts: Start by reading the card and getting the companion runtime working on a harmless test task. Do not skip it, because the card says the model and runtime are designed to be used together. The card says the family learned from long trajectories with dependent steps and outcome checks. Training used staged learning, with grounding on SFT checkpoints and task completion on later joint desktop and web RL checkpoints. The 3B active figure reflects work done per token, so loading the full weights still takes far more memory. ```flow User request and screenshot go in Model reasons about the screen It returns a tool call, 0 to 1000 coordinates Runtime converts to pixels and acts A new screenshot comes back ``` Good to know: - License: Apache 2.0 - Cost to run: none named, though the weights are about 70.2 GB in BF16 - Data it reaches: screenshots and interaction history of the desktop or browser it controls, plus Hugging Face for the download - Not for: use without the official companion runtime, which the card calls required for computer use - Limits: the card says its reported results come from the technical report and are not new measurements of these files ```chart Title: NorthCUA-Mini reported scores in percent ScreenSpot-Pro: 92.41% OSWorld-G: 87.94% OSWorld-Verified: 85.30% WebArena-Verified: 70.43% ``` ## On Hugging Face, 13 November 2026 - Task: image text to text - Library: transformers - Parameters: 35.1B - Weights: about 70.2 GB in BF16; about 17.6 GB at 4-bit (estimate) - Formats: Safetensors - Downloads in the last 30 days: 739 - Likes: 17 - License: apache-2.0 - Quantized versions: 4 - Tags: transformers, safetensors, qwen3_5_moe, image-text-to-text, northcua, computer-use, multimodal, conversational ## Links - Model: https://huggingface.co/beipo-ai/NorthCUA-Mini - Files: https://huggingface.co/beipo-ai/NorthCUA-Mini/tree/main ## Get the model Download its files with Hugging Face's command line: ```sh pip install -U huggingface_hub hf download beipo-ai/NorthCUA-Mini ``` 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 [beipo-ai/NorthCUA-Mini on Hugging Face](https://huggingface.co/beipo-ai/NorthCUA-Mini) ## 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 | | --- | --- | --- | --- | --- | --- | | NorthCUA-Mini (this page) | 739 | 17 | 35.1B | apache-2.0 | 2026-11-03 | | [NorthCUA-Pro](https://mindket.com/p/f9762fb8-e2e3-411e-a85e-4ca81412ffd3) | 1,066 | 33 | 397B | apache-2.0 | 2026-11-03 | | [NorthCUA-Nano](https://mindket.com/p/fe172c1f-6c3c-40c1-ab80-6426435c7406) | 485 | 14 | 9.4B | apache-2.0 | 2026-11-03 | | [Holo4-27B](https://mindket.com/p/139fcb8f-6383-4be5-b579-d9cfce8ce9d3) | 2,486 | 105 | 27.4B | cc-by-nc-4.0 | 2026-09-28 | | [Fara1.5-9B](https://mindket.com/p/93437033-da29-4dfb-8cfb-5400e03005ae) | 3,008 | 64 | 9.4B | mit | 2026-07-27 | | [Mage-VL-Cowork](https://mindket.com/p/53460bdf-08aa-4019-bb00-2a650a2d0157) | 4,226 | 222 | 4.7B | apache-2.0 | 2026-10-30 | 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: 2026-11-03 - First published: 2026-11-01 - Checked on Hugging Face: 2026-11-12 - Something wrong? Tell us: https://mindket.com/takedown#ask ## Get it - [Open beipo-ai/NorthCUA-Mini on Hugging Face](https://huggingface.co/beipo-ai/NorthCUA-Mini) - [Get it on your Mindket account](https://account.mindket.com/account/?product=mindket#/market/50dfec70-c66f-4693-9f0a-887c6469974b) - [Its page on Mindket](https://mindket.com/p/50dfec70-c66f-4693-9f0a-887c6469974b)