OpenAI Launches Open Source GPT Models on Azure and Windows AI Platforms

OpenAI has unveiled its gpt-oss models, marking its first open-weight release since GPT-2. These models, gpt-oss-120b and gpt-oss-20b, are now available on Azure AI Foundry and Windows AI Foundry, offering developers and enterprises unprecedented control over AI deployment and customization.

Key Highlights:

• gpt-oss-120b: A 120-billion-parameter model optimized for complex reasoning tasks like math and code, capable of running on a single enterprise GPU.
• gpt-oss-20b: A lightweight model designed for agentic tasks, suitable for local deployment on Windows devices with 16GB+ VRAM.

Azure AI Foundry and Windows AI Foundry Integration

Microsoft's Azure AI Foundry provides a unified platform for fine-tuning, deploying, and managing AI models, while Windows AI Foundry extends this capability to edge devices. Key features include:

• Fine-tuning: Use parameter-efficient methods like LoRA, QLoRA, and PEFT to adapt models for specific domains.
• Edge Deployment: Optimize models for offline inference on devices via Foundry Local.
• Hybrid AI: Mix open and proprietary models to balance performance, cost, and compliance.

Developer and Enterprise Benefits

• Transparency: Full access to model weights for inspection and customization.
• Flexibility: Deploy models in the cloud or locally, with options for secure, low-latency inference.
• Cost Efficiency: Competitive pricing for managed compute resources.

Pricing and Availability

• gpt-oss-120b and gpt-oss-20b are available now on Azure AI Model Catalog.
• Foundry Local supports gpt-oss-20b on Windows, with macOS compatibility coming soon. QuickStart guide.