How to Run OpenAI’s New GPT Model Locally on Windows with Microsoft’s AI Foundry

Running a powerful language model directly on your own PC used to sound like something only researchers or AI labs could do. But with OpenAI’s new gpt-oss-20b model and Microsoft’s support through Windows AI Foundry, it’s now possible to bring local AI to your desktop—if you’ve got the hardware for it.

Here’s how it works, and what you’ll need to get started.

What Is GPT-OSS-20B?

Let’s start with the basics. gpt-oss-20b is OpenAI’s latest open-weight model, released for free and optimized for running locally. It’s designed to handle code execution, tool use, and autonomous workflows, so it's a bit different from models made just for chatting.

Microsoft has jumped in quickly to make this model easily available through its AI Foundry on Windows, which streamlines the setup process and handles a lot of the heavy lifting under the hood.

What You’ll Need

Before anything else, you’ll want to make sure your machine is actually up for the job. Here’s what’s required:

• A Windows PC or laptop
• At least 16GB of dedicated VRAM (from an Nvidia RTX 30/40 series or a high-end AMD Radeon GPU)
• Enough SSD space to host the model (expect multiple GBs)
• Updated drivers and access to Microsoft’s AI Foundry tools

If you’re using a gaming or workstation PC from the last couple of years, you might already meet the GPU requirement. If not, this isn’t going to work on your everyday Ultrabook or budget laptop.

Step 1: Get the Windows AI Foundry Toolkit

Start by downloading the Windows AI Foundry tools. Microsoft hasn’t rolled this out as a mainstream feature just yet, so you’ll likely find it through official dev channels or GitHub.

Once installed, AI Foundry helps manage local models and supports hardware acceleration so you’re not relying entirely on cloud inference.

Step 2: Download GPT-OSS-20B

Inside the Foundry interface or via command line tools, you can now download and initialize the gpt-oss-20b model.

Microsoft has already pre-optimized the model for Windows, so it should run more efficiently than trying to set it up manually. Expect it to take a while depending on your internet speed and storage.

Step 3: Start Running It Locally

With everything in place, you can now run GPT-OSS-20B directly on your system. Whether you're integrating it into a tool you’re building or just testing it through the terminal, it’ll be using your local compute power—not the cloud.

Microsoft says it’s a great fit for low-bandwidth environments, developers building autonomous agents, or just anyone who wants more control over their data and AI workflows.

macOS Support Is Coming Soon

While the current release is Windows-focused, Microsoft confirmed that macOS support is on the way. There’s no exact timeline, but it looks like Apple Silicon users may soon be able to try local GPT as well.