OpenAI has introduced a new free and open GPT model that is compatible with personal computers, with Microsoft facilitating its use for Windows users. The lightweight gpt-oss-20b model is now accessible via Windows AI Foundry, and a version for macOS is anticipated to launch soon.
To successfully run this model, users need a PC or laptop equipped with a minimum of 16GB of VRAM. This means having one of the leading GPUs from Nvidia or certain Radeon GPUs with adequate VRAM. The gpt-oss-20b model has been designed specifically for optimized code execution and tool integration. According to Microsoft, it is “ideal for creating autonomous assistants or incorporating AI into practical workflows, even in environments with bandwidth limitations.”
Microsoft has optimized the gpt-oss-20b model for local inference and hinted at the potential for broader device support in the near future. This could suggest that a more tailored version for Copilot Plus PCs may emerge, similar to the recent variety of local AI models that Microsoft has incorporated into Windows.
The swift integration of OpenAI’s latest model into Windows AI Foundry by Microsoft coincides with Amazon’s rapid adoption of the new open-weight GPT-OSS models for its cloud computing services. This marks a significant moment, as it is the first instance of running an OpenAI model locally on a Windows platform, while concurrently giving Microsoft’s primary cloud competitor access to cutting-edge OpenAI models, thereby adding a new layer of complexity to the existing relationship between OpenAI and Microsoft.