Microsoft Expands AI Access with GPT-OSS Models

Prime Highlights

Microsoft launches OpenAI’s first open-weight models since GPT-2, offering gpt-oss-120B and gpt-oss-20B.
Models now available for cloud deployment via Azure AI Foundry and local deployment through Windows AI Foundry.

Key Fact

The gpt-oss-20B model can run locally on Windows PCs with 16 GB or more GPU memory.
The gpt-oss-120B model handles high-complexity reasoning on a single datacenter-class GPU.

Key Background

OpenAI published its first open-weight models in years, which was a significant step towards AI democratization. OpenAI introduced the gpt-oss family of new models in two variants: large gpt-oss-120B and small gpt-oss-20B. Releasing the model weights to the world enables developers to fine-tune, experiment with, and customize these models for specialized use cases, offering a new level of flexibility and openness.

The gpt-oss-120B boasts a monumental 120-billion-parameter model, sparsified to enable dense reason-intensive work. A pro at deep problem-solving, sophisticated coding issues, and exploration by topic—while still digestible on a single enterprise-grade GPU. The gpt-oss-20B is, instead, optimized for responsiveness, particularly for agentic use cases like using software tools, automation, and integrated development environments. This lighter alternative can be implemented in-house on current Windows operating systems with 16 GB of VRAM or higher, and therefore is accessible to a greater community of developers.

These models can be scaled by organizations and execute them in the cloud on Azure AI Foundry. The platform supports simple evaluation, tuning, governance, and endpoint configuration of AI workloads. Enterprises can combine gpt-oss with other open-source or custom models for differently varied operational requirements, all within a secure, managed environment.

For device AI, Windows AI Foundry provides developers with the ability to execute models offline completely. With Local SDKs and tools in Foundry, the development teams can create and test AI applications on Windows devices natively with privacy intact and latency minimized by removing the constant need for cloud connectivity.

Together, Azure AI Foundry and Windows AI Foundry form a hybrid AI platform. They offer the organizations and developers flexibility to run AI applications where it will best serve—cloud for scalability, edge for control and privacy—or to achieve the best of both worlds with the most open and flexible models from OpenAI.