For the first time, Microsoft has included OpenAI’s free and open-source GPT model, gpt-oss-20b, into Windows 11, allowing users to run sophisticated AI features directly on their devices. The integration represents a significant move toward on-device artificial intelligence processing and was revealed Tuesday via Microsoft’s Windows AI Foundry platform.
Computers with at least 16GB of VRAM—typically found in contemporary Nvidia or AMD Radeon graphics cards—are required for the gpt-oss-20b model. Microsoft claims that because the model is “tool-savvy and lightweight,” it is perfect for activities like code execution and tool usage. This makes it perfect for creating self-sufficient assistants or incorporating AI into workflows, even in settings with spotty internet access.
Windows’s First Local OpenAI Model
In contrast to the cloud-dependent strategy that has dominated AI services, this is the first time an OpenAI model can operate locally on Windows devices. In order to ensure effective performance on consumer hardware, the model, which was released on Tuesday, uses a mixture-of-experts architecture with 21 billion parameters, of which only 3.6 billion are active each job.
Microsoft claims in a blog post that it envisions “a future where AI is ubiquitous—and we are committed to being an open platform to bring these innovative technologies to our customers, across all our data centers and devices” and described the integration as part of a “bigger story” toward ubiquitous AI.
Performance And Restrictions
High-compute reinforcement learning was used by OpenAI to train gpt-oss-20b, putting it in a position to perform well when executing tools like Python code or online searches and powering AI agents. The model does have some significant drawbacks, though, such as a 53% hallucination rate on OpenAI’s PersonQA benchmark and text-only functionality devoid of visual or audio processing.
Plans For Expansion
Microsoft has already made the gpt-oss-20b model available through its Azure AI Foundry cloud platform and intends to shortly expand its availability to macOS users. For the first time, Microsoft’s main cloud rival now has access to the most recent OpenAI models thanks to the model’s availability on Amazon Web Services.
The model is presently only supported by Nvidia GPUs, although Windows 11 users can access it using Microsoft’s Foundry Local tool or the AI Toolkit for Visual Studio Code. Local deployment across various hardware configurations is also made possible by alternative methods such as LM Studio and Ollama.