Microsoft recently announced that through its Windows AI Foundry platform, it is now offering the latest free open-source large model from OpenAI, gpt-oss-20b, to Windows 11 users. This means users can directly access powerful AI features and various popular open-source models on their local computers without relying on the cloud.
Microsoft noted in its blog that gpt-oss-20b is a lightweight and efficient model, especially good at performing tasks such as executing code and calling external tools. It can run efficiently on various Windows hardware and will support more devices in the future. Even in environments with limited network bandwidth, this model is suitable for building autonomous AI assistants or integrating AI into daily workflows.
According to the information, this model can run on mainstream consumer PCs or laptops with at least 16GB of VRAM. OpenAI stated that gpt-oss-20b has been trained using intensive computing resources, particularly good at handling "chain-of-thought" tasks, such as calling tools for web searches or executing code.
However, as OpenAI's "smallest" open-source model, gpt-oss-20b only supports text processing and cannot generate images or audio. OpenAI also warned that this model has a relatively high "hallucination" rate, with approximately 53% of its answers containing factual errors in internal testing.
Other than Windows 11, Microsoft said it plans to introduce the model to more platforms such as macOS in the future. Currently, gpt-oss-20b is available on Microsoft's Azure AI Foundry and Amazon's AWS platforms, providing more choices for cloud developers.