OpenAI o1 & OpenAI o3-mini on Foundry — Azure AI Foundry Architecture & Agent Service

The Latest OpenAI Models on Azure

OpenAI o1 & o3-mini Reasoning Models on Azure AI Foundry

OpenAI o1 and o3-mini established Azure's reasoning tier on Microsoft Foundry, now expanded in 2026 with OpenAI o3, o3-pro, o4-mini, GPT-6 Astra (1.05M context), and GPT-6.1 Sol. Reasoning models feature configurable reasoning_effort (low, medium, high — or reasoning={"effort": "high"} in the OpenAI Responses API) for complex STEM, coding, and agentic planning workloads.

OpenAI o3-mini - High-Efficiency Reasoning

The OpenAI o3-mini family brings native structured outputs, function/tool calling, and developer messages combined with profound system-2 reasoning capabilities at low latency and cost.

Key Features Across Azure OpenAI Models

• 1 Million Token Context Window (GPT-4.1 Series): Process entire repositories or massive document sets at once with GPT-4.1, GPT-4.1 Mini, and GPT-4.1 Nano.
• Configurable Reasoning Effort (o1 / o3-mini): Tune internal chain-of-thought compute dynamically per request using reasoning_effort: "low" | "medium" | "high".
• Dynamic Tool Calling & Structured Outputs: Strict JSON schema adherence across both reasoning (o1, o3-mini) and multimodal (GPT-4.1, GPT-4o) deployments.

Azure-Specific Deployments

Unlike the public OpenAI API, deploying OpenAI o1, o3-mini, and GPT-4.1 on Azure AI Foundry provides:

• VNet Integration: End-to-end private networking via Azure Private Link. Your prompts never traverse the public internet.
• Provisioned Throughput (PTU): Reserve dedicated capacity so your latency remains stable even during peak global usage.
• Integrated PII Redaction: Combine with Azure AI Content Safety and native PII detectors to scrub sensitive data before the prompt reaches the model.