Microsoft Foundry’s GPT-5.6 and Hosted Agents: Building Production-Grade Agent Systems Across Regions
Discover how Microsoft’s latest updates enhance AI systems with GPT-5.6 and hosted agents, ensuring reliability and efficiency across regions.
Microsoft has moved GPT-5.6 and production agent capabilities into general availability within Microsoft Foundry, alongside an Asia-Pacific Data Zone and a hosted-agent runtime in the Foundry Agent Service. For teams building cloud-native AI systems, this is more than a new model drop: it’s an update to the production control plane around agent workloads—covering where models run, how agents execute reliably, how they access enterprise knowledge and tools, and how organizations measure quality and cost after deployment.
This matters because agentic applications fail in ways that chatbots don’t: long-running workflows, tool calls, partial failures, identity-bound access, and real operational monitoring. With GPT-5.6 availability, regional deployment options, and hosted agents, Foundry shifts agent development from
Frequently Asked Questions
What’s the difference between using GPT-5.6 in Foundry versus just calling a model in an app?
In Foundry, GPT-5.6 is paired with production agent capabilities and a control plane for agent workloads. That means the focus is not only model quality, but also where the model runs, how long-running agent workflows execute, how tool calls are managed, and how identity-bound access and monitoring work across environments.
How do Hosted Agents in Microsoft Foundry help with reliability issues that chatbots don’t face?
Agent systems often include long-running workflows, repeated tool calls, and partial failures. Hosted agents are designed to run reliably in the Foundry Agent Service so teams can better handle retries, intermediate states, and orchestration. This reduces the operational burden compared to building reliability logic manually in each application.
What does the Asia-Pacific Data Zone change for teams deploying agent workloads?
The Asia-Pacific Data Zone is about regional placement of data and workloads, which can matter for latency, data residency, and governance. By aligning agent execution with the selected regional zone, organizations can simplify compliance requirements while keeping performance predictable for users and dependent systems in the region.
How does Foundry’s production control plane affect access to enterprise knowledge and tools?
A production control plane focuses on how agents access enterprise resources reliably and safely. In Foundry, agents are built to work with enterprise knowledge and tool integrations while respecting identity-bound access. This helps teams avoid ad-hoc permissions in application code and supports consistent enforcement across agent deployments.
After deployment, how can organizations measure quality and cost for agent systems in Foundry?
Unlike simple chat interactions, agents involve multi-step execution, tool calls, and varying failure modes. Foundry’s production agent setup is intended to support ongoing measurement of quality and cost post-deployment. Teams can evaluate whether workflows succeed, how often tools are invoked, and the resulting resource usage to optimize operations over time.