Microsoft brings OpenAI’s smallest open model to Windows users

Microsoft is set to enhance the capabilities of Windows 11 users by introducing OpenAI’s latest free and open model, gpt-oss-20b, through its Windows AI Foundry platform. This initiative allows users to utilize AI features, APIs, and a variety of popular open-source models directly on their computers. In a recent blog update, Microsoft described gpt-oss-20b as "tool-savvy and lightweight," emphasizing its optimization for agentic tasks, including code execution and tool utilization. The model operates efficiently across various Windows hardware configurations, with plans to support even more devices in the near future. It is particularly well-suited for developing autonomous assistants or integrating AI into real-world applications, even in settings with limited bandwidth. Launched recently, OpenAI’s gpt-oss-20b is compatible with consumer PCs and laptops that possess a minimum of 16GB of VRAM, typically found in modern GPUs from Nvidia or Radeon. The training process for this model involved high-compute reinforcement learning, enabling it to effectively power AI agents and perform tasks such as web searches and Python code execution as part of its reasoning process. However, it is important to note that gpt-oss-20b is limited to text-only outputs, which means it cannot generate or process images and audio, unlike some of OpenAI's other offerings. Additionally, the model has shown a tendency to produce inaccurate information, as evidenced by its performance on OpenAI’s PersonQA benchmark, where it hallucinated in 53% of the tested queries. Microsoft has announced plans to extend the availability of gpt-oss-20b to macOS users, along with support for additional devices, though specific details have not been disclosed. Furthermore, both gpt-oss-20b and the previous model, gpt-oss-120b, will also be accessible through Microsoft’s hosted Azure AI Foundry platform, as well as on Amazon’s AWS.