OpenAI's New Open Models Now on Microsoft Azure AI Foundry
The artificial intelligence landscape is witnessing a significant shift as OpenAI’s newly released open-weight models, gpt-oss-120b and gpt-oss-20b, are now accessible on Microsoft’s Azure AI Foundry. This strategic integration empowers developers with unprecedented flexibility to build, test, and deploy AI-powered applications, marking a pivotal moment in the democratization of advanced AI.
OpenAI’s gpt-oss models, introduced on August 5, 2025, represent the company’s first open-weight release since GPT-2. Unlike previous proprietary models, these models are licensed under the permissive Apache 2.0 license, allowing for free use, modification, and redistribution, even for commercial purposes. This move is seen as a strategic play by OpenAI to broaden enterprise reach and challenge competitors like Meta in the open-source AI space.
The gpt-oss family includes two main models: gpt-oss-120b, a larger model with 117 billion total parameters and 5.1 billion active parameters per token, and gpt-oss-20b, a more compact version with 21 billion total parameters and 3.6 billion active parameters. Both models leverage a Mixture-of-Experts (MoE) architecture and 4-bit quantization (MXFP4), which significantly enhances inference speed and reduces resource consumption. The gpt-oss-120b model can operate on a single enterprise GPU, while the gpt-oss-20b is optimized for consumer hardware, running efficiently with just 16GB of memory. These models are particularly strong in reasoning, coding, and tool-use tasks, demonstrating capabilities that rival OpenAI’s own proprietary models like o4-mini on core benchmarks.
Microsoft’s Azure AI Foundry serves as the crucial platform for these models, offering a unified environment for the entire AI development lifecycle, from initial exploration to full-scale production. The Foundry provides developers with a comprehensive toolkit, including pre-built models, customizable templates, and scalable infrastructure. Its integration with familiar development environments like GitHub and Visual Studio, alongside robust security and compliance features, makes it an attractive choice for enterprises. Developers can fine-tune the gpt-oss models with their proprietary data, optimize them for edge devices, and deploy them in containerized environments, accelerating the journey from concept to deployment. This means custom AI solutions, or “checkpoints,” can be shipped in hours rather than weeks.
The availability of gpt-oss models on Azure AI Foundry underscores a broader industry trend towards open-weight and open-source AI models. This shift democratizes AI access, lowers barriers for emerging markets and smaller organizations, and fosters innovation by allowing developers to customize and build upon existing models without the constraints of proprietary systems. Enterprises are increasingly embracing open-source AI solutions due to their cost-effectiveness, flexibility, and transparency. While Microsoft maintains exclusive rights to OpenAI’s proprietary API on Azure, the Apache 2.0 license for gpt-oss models allows other cloud providers, like Amazon Web Services, to offer them as well, intensifying competition in the cloud AI market.
This collaboration between OpenAI and Microsoft, particularly with the latest release of GPT-5 being incorporated into Microsoft’s offerings, reinforces Azure’s position as a leading platform for cutting-edge AI development. The focus on open-weight models within Azure AI Foundry provides developers with greater control, lower latency, and enhanced privacy for their AI applications, particularly for on-device inference and agentic tasks.