The GPT-5 family of models is now available in Azure AI Foundry Agent Service, bringing Azure OpenAI’s most advanced reasoning, multimodal, and coding intelligence into a platform purpose-built for enterprise-scale agent development.
The new generation of models
With this release, developers can choose from the full GPT-5 lineup to balance speed, depth, and cost for their agent development needs:
- GPT-5 — the flagship, with a 272k-token context window, designed for deep analysis, complex automation, and high-trust scenarios such as analytics and compliance.
- GPT-5-mini — fast and efficient, ideal for real-time interactions and reliable tool use.
- GPT-5-nano — ultra-low latency and cost-optimized for high-volume requests and lightweight orchestration.
- GPT-5-chat — a multimodal specialist with a 128k-token context window, enabling natural conversation and contextual reasoning across documents and images.
These models join additional Azure OpenAI families such as o4-mini and o3, giving developers a broad toolkit to scale from simple Q&A to advanced multi-agent orchestration.
Features built for enterprise value
Foundry Agent Service goes beyond raw model access—it turns GPT-5 into production-ready agents with capabilities that deliver business impact:
- Streaming responses for interactive, real-time engagement.
- Flexible tool calling so agents can connect to APIs, databases, and systems—whether with structured queries or free-form inputs like SQL and scripts.
- Structured outputs for predictable, typed responses that integrate cleanly with downstream workflows.
- Multimodality, allowing agents to read documents, interpret charts, and combine visual and textual reasoning.
- File Search and Code Interpreter for grounded retrieval and safe, on-demand computation.
- Model routing that automatically selects the best GPT-5 variant for each task, balancing performance, accuracy, and cost.
Enterprise readiness at GA
With GPT-5 running inside Foundry, enterprises now have generally available capabilities to develop agents securely and at scale:
- Trust and governance — apply Azure RBAC, monitor usage and cost, filter content, and enforce compliance requirements.
- Bring-your-own resources — run in private VNets, store threads in customer-owned Cosmos DB, and maintain residency and retention control.
- Multi-agent workflows — coordinate specialized agents for onboarding, logistics, finance, or creative work.
- Open standards — integrate with Model Context Protocol (MCP) and A2A to ensure interoperability without lock-in.
- AgentOps — trace and evaluate agent behavior, monitor with dashboards, and fine-tune continuously.
In short, Foundry ensures organizations can move from proof-of-concept to mission-critical AI, with reliability, auditability, and scale.
Real-world examples
Imagine an insurance claims assistant where GPT-5 analyzes documents, calls fraud detection APIs, and produces compliant summaries for reviewers—with every step logged and auditable. Or a supply chain agent that merges product Q&A, order resolution, and logistics troubleshooting—switching dynamically between GPT-5 variants to optimize speed or reasoning depth. These are the workflows Foundry enterprise agents are designed to run in production, starting today.
Availability and what’s next
- All GPT-5 models (plus expanded o-series support), including File Search and Code Interpreter, are available now via SDK, API and the Agents Playground in Foundry Developer Portal.
- GPT-5 registration is required: Access will be granted according to Microsoft's eligibility criteria. If you have already requested access, thank you for your patience while your submission is reviewed. Request access
- Microsoft tool integrations, such as SharePoint and Bing, are coming next, connecting agents directly into knowledge and productivity sources.
Start building today
- Check model availability by region in the model support documentation.
- Get hands-on: Quickstart: Create and deploy an agent - Microsoft Copilot Studio
Microsoft