Microsoft has expanded its Azure AI Foundry and Windows AI Foundry platforms with OpenAI’s gpt-oss open-weight language models, gpt-oss-120b and gpt-oss-20b, to provide developers with greater flexibility and control over AI implementation.
The introduction of these models is significant as it caters to different needs. The gpt-oss-120b model is designed for high-performance reasoning tasks. On the other hand, the more compact gpt-oss-20b model is capable of functioning on PCs equipped with GPUs featuring at least 16GB of memory. By granting developers full access to the models’ weights, Microsoft enables them to fine-tune the models for specific applications, use them offline, or create industry-specific assistants.
This open-weight access also facilitates model review, partial retraining, and export for deployment on Microsoft’s Azure Kubernetes Service (AKS) or local machines. Developers can leverage these features to customize the models according to their requirements. Azure AI Foundry supports these models with a range of tools for evaluation, fine-tuning, and deployment. The platform boasts a catalog of over 11,000 models, providing developers with a vast resource for their AI projects.
Additionally, Foundry Local provides on-device support for local inference, catering to needs that require enhanced security or offline capabilities. The gpt-oss-20b model is currently available on Windows, with macOS compatibility planned for the near future. Both models will be integrated with the common responses API, ensuring seamless interaction with other Microsoft services.
Microsoft emphasizes that this initiative offers businesses and developers more transparency and options for managing AI across diverse environments, including cloud, on-device, and edge computing. By providing open-weight language models and supporting tools, Microsoft aims to empower developers to harness the potential of AI in their applications and services.
