Microsoft has announced the integration of OpenAI’s latest open-source language model, gpt-oss-20b, into Windows 11 through its Windows AI Foundry platform, enabling users to access powerful AI capabilities locally on their PCs.

Described by Microsoft as “tool-savvy and lightweight,” the gpt-oss-20b model is designed for tasks such as code execution and tool use, making it ideal for developers building autonomous agents or integrating AI into real-world workflows.

The model runs efficiently on consumer-grade hardware, specifically PCs and laptops with at least 16GB of VRAM, typically found in modern Nvidia or AMD Radeon GPUs.

Although text-only, meaning it doesn’t support image or audio processing like some of OpenAI’s larger models, gpt-oss-20b is optimized for agentic tasks such as web search and executing Python code as part of its reasoning process.

OpenAI says it trained the model using high-compute reinforcement learning, boosting its capabilities in tool use and step-by-step logic.

However, gpt-oss-20b is not without limitations. According to OpenAI, it tends to hallucinate — or generate false or misleading information — in response to a high number of queries. On OpenAI’s internal PersonQA benchmark, the model hallucinated answers to 53% of questions, raising concerns about its reliability for factual tasks.

Microsoft confirmed in a blog post that it will expand support for the model to macOS and other platforms, though it did not specify a timeline. In addition to Windows, both gpt-oss-20b and the larger gpt-oss-120b model are available through Microsoft’s Azure AI Foundry and on Amazon Web Services.

The move underscores Microsoft’s continued investment in democratizing AI development by providing access to open, local models for developers and power users alike.