Frontier models and production agents: Advancing Microsoft Foundry for the agentic era

Microsoft Foundry is now generally available with significant new capabilities, including the integration of GPT-5.6 models, the launch of an Asia-Pacific Data Zone, and the widespread availability of hosted agents within the Foundry Agent Service. These advancements signal a pivotal moment in the maturation of AI agent development, moving the technology from experimentation to robust, enterprise-grade production environments. The platform aims to democratize AI agent creation, allowing developers to build, deploy, and manage sophisticated AI agents seamlessly within their existing workflows and infrastructure.

The drive towards agentic AI, where AI systems can proactively perform tasks and interact with the digital world, has been a major focus for Microsoft. The company’s commitment to making AI accessible and valuable in real-world systems—ones that are reliable, observable, and directly contribute to business objectives—is underscored by these latest releases. With over 100,000 organizations already leveraging Microsoft Foundry, and prominent companies like Adobe, Telefónica, and Tata Consultancy Services actively deploying agents in production, the platform’s growing adoption highlights its significance in the current AI landscape.

At the recent Microsoft Build conference, Microsoft articulated a clear vision for the agentic era: empowering developers to build agents where they already work, run them on trusted infrastructure, and easily distribute them to end-users. This vision is now being realized with three key updates that are now generally available within Microsoft Foundry. These updates collectively bring together advanced AI models, production-grade agent runtimes, robust identity and security controls, and seamless distribution across Microsoft 365, all within a unified platform. This integrated approach aims to eliminate the friction organizations often face when attempting to move AI initiatives from pilot phases to full-scale production by reducing the need to cobble together disparate tools and services.

Why Foundry Stands Out as the Premier Agent Platform

Microsoft Foundry is engineered as a comprehensive, end-to-end platform for the entire lifecycle of AI agents. It encompasses building, running, governing, and distributing these intelligent agents. The platform’s strength lies in its ability to consolidate critical capabilities across three foundational pillars:

• Model Access and Flexibility: Providing access to a diverse range of cutting-edge AI models, from proprietary frontier models to open-source options, allowing organizations to select the best model for specific tasks.
• Production-Ready Runtime and Infrastructure: Offering a stable and scalable environment for agents to operate, complete with essential features like memory, tool integration, and real-time event processing.
• Enterprise-Grade Governance and Distribution: Ensuring that agents can be deployed with robust security, compliance, and identity management, and can be easily distributed to users across various Microsoft ecosystems.

These pillars are brought to life by the latest Foundry updates, which are designed to simplify and accelerate the process of building, running, and scaling production-ready AI agents on a single, cohesive platform.

Building with Any Framework and Model on an Industry-Leading AI Platform

Agent development is increasingly integrated into existing developer workflows. Foundry facilitates this by enabling development directly within tools like GitHub Copilot and Microsoft Visual Studio Code (VS Code). The Foundry Toolkit for VS Code and the Foundry skill abstraction layer simplify the deployment process to Foundry. Whether developers are utilizing the Microsoft Agent Framework, the GitHub Copilot SDK (now generally available), or the Claude Agent SDK, Foundry serves as the ultimate production destination. The journey begins with selecting the appropriate AI model for the task at hand.

Selecting the Right Model for Every Task

The efficacy of an AI agent is intrinsically linked to the intelligence and reasoning capabilities of its underlying model. Microsoft Foundry provides organizations with unified access to a spectrum of industry-leading models, including frontier AI models, open-source alternatives, and specialized task-specific models. This curated selection empowers teams to precisely match the model’s capabilities to the demands of each workload, rather than being constrained to a single, generalized model.

A significant development in this release is the general availability of OpenAI’s GPT-5.6 series within Microsoft Foundry Models and the Microsoft Foundry Agent Service. This series introduces three distinct models:

• GPT-5.6 Sol: Designed for complex reasoning and nuanced tasks, offering the highest level of capability.
• GPT-5.6 Terra: Strikes a balance between advanced reasoning and efficient performance, suitable for a wide range of demanding applications.
• GPT-5.6 Luna: Optimized for speed and cost-effectiveness, ideal for high-volume, less complex tasks.

This tiered approach provides organizations with the crucial flexibility to align model selection with specific business requirements concerning capability, cost, and performance. Instead of forcing every application onto a single, potentially over- or under-powered model, businesses can now strategically choose the most appropriate GPT-5.6 variant for each scenario.

Customer feedback consistently emphasizes that access to the latest AI models is as critical as the quality of those models themselves. To address this, Microsoft is making the GPT-5.6 series available across its global infrastructure from day one. This includes availability in:

• Global Standard and Global Priority Processing: Across all 28 existing global regions, ensuring low latency and high availability.
• Data Zones Standard: For regions requiring localized data processing.
• Global Provisioned: Offering dedicated capacity for predictable performance and scalability.

This broad deployment strategy ensures that customers can rapidly adopt the newest frontier AI innovations within their existing, established application footprints, regardless of their geographical location or specific data residency requirements.

GPT-5.6 Pricing Structure

The introduction of the GPT-5.6 series comes with a transparent and tiered pricing model designed to accommodate varying usage patterns and budgets:

Note: Pricing for Data Zone and Provisioned deployments may vary.

Running Frontier AI Where Your Business Operates

Expanding access to advanced models is only one part of platform evolution; enabling their compliant execution in diverse geographical locations is equally vital. The newly launched Asia-Pacific (APAC) Data Zone for Microsoft Foundry directly addresses this need. This feature provides APAC customers with the capability to run advanced OpenAI models while ensuring that data processing remains within the Asia-Pacific regions. This eliminates the complexity of managing separate environments or waiting for regional capabilities to catch up, offering a streamlined and compliant solution for AI adoption in the region.

With Foundry’s comprehensive deployment options—including Global, Data Zone, and Regional configurations—organizations can now align their AI strategies with specific sovereignty, compliance, performance, and scalability mandates. Crucially, this is achieved while maintaining a consistent and familiar development and operational experience across all environments.

Hongsoo Kim, Chief Data and AI Officer (CDAO) at Viva Republica (Toss), commented on the significance of these regional capabilities: "As financial institutions adopt AI, responsible data handling becomes foundational to trust. Microsoft Foundry’s APAC Data Zone allows us to keep data processing regionally anchored while accessing advanced AI models at scale. This gives us the confidence to accelerate AI innovation responsibly and reinforces our ambition to be a leading AI-powered financial platform in Asia."

Generating Impact with Action-Oriented, Context-Aware Agents

A powerful model is merely the foundation for an effective AI agent. To transition from potential to production, an agent requires a robust runtime environment. This includes a reliable place to execute, an understanding of business context, governed access to necessary tools, persistent memory across interactions, the ability to act on real-world events, and a clear pathway to the end-users who will benefit from its capabilities. Foundry integrates each of these essential components as built-in features, designed to work harmoniously.

The platform provides:

• Hosted Agents: A managed service for deploying and running agents, simplifying infrastructure management.
• Agent Tooling: A framework for defining and integrating external tools and services that agents can leverage to perform actions.
• Contextual Memory: Enabling agents to retain information and learn from past interactions, leading to more personalized and efficient engagements.
• Event-Driven Actions: Allowing agents to respond to and trigger actions based on real-world events, facilitating dynamic and proactive workflows.
• Microsoft 365 Distribution: Seamless integration with Microsoft 365 applications, enabling agents to reach users within their familiar productivity environment.

Governing and Optimizing the Full AI Lifecycle with Observability and Controls

An AI agent that cannot be monitored, improved, or secured is an agent that cannot be trusted in a production setting. Microsoft Foundry prioritizes trust as a core platform tenet, rather than placing the burden solely on developers. The latest release closes the loop on the post-development lifecycle by providing comprehensive tools for observing agent performance, enabling continuous improvement, and validating their value.

Key advancements in governance and optimization include:

• Agent Observability: Detailed logs and telemetry provide insights into agent behavior, decision-making processes, and resource utilization, allowing for rapid troubleshooting and performance analysis.
• Continuous Improvement Tools: Features like prompt optimization and agent tuning help refine agent responses and actions over time, leading to enhanced accuracy and efficiency.
• Security and Compliance: Robust controls ensure that agents adhere to enterprise security policies and regulatory requirements, safeguarding sensitive data and operations.
• Cost Management and Optimization: As agents scale from limited pilots to thousands of daily operations, Foundry equips teams with the necessary levers to manage and predict expenditure without leaving the platform.

This cost management is facilitated by a multi-faceted approach that begins with choice:

• Model Choice: Selecting the most cost-effective model that meets performance needs for each specific task.
• Prompt Caching: Reducing redundant computations by storing and reusing results of frequently encountered prompts.
• PTU Spillover and Quota Optimization: Intelligent traffic management mechanisms that ensure service continuity during peak loads and optimize the utilization of provisioned throughput units (PTUs).

Building upon this foundation, the model router intelligently directs each request to the most appropriate model. Prompt caching minimizes repetitive computations, while PTU spillover and quota optimization maintain service continuity during usage spikes. For agents, toolboxes in Foundry ensure that only the necessary tools are invoked for a given request, and the agent optimizer refines prompts, skills, tools, and model choices based on custom evaluators, ensuring peak performance and efficiency.

Beyond cost savings, Foundry provides a clear view of ROI for agents. This feature integrates business value, usage metrics, and operational costs into a single dashboard, enabling teams to assess whether production agents are generating more value than they incur in costs, and to identify areas where cost may be outpacing value.

For a practical demonstration of these capabilities, a new Microsoft Mechanics episode titled "Token Economics for Agents" offers an in-depth walkthrough.

In Production: Real-World Applications Built on Foundry

The organizations adopting Microsoft Foundry are not merely experimenting; they are actively deploying sophisticated AI solutions. This includes a diverse range of entities, from digital-native startups to some of the world’s largest enterprises.

Case studies and testimonials highlight a consistent pattern: teams that previously spent weeks integrating, securing, and deploying AI agents can now accomplish these tasks in a matter of days. They are leveraging infrastructure that meets stringent compliance standards and delivering AI-powered capabilities to users through the tools they already use daily.

Examples of production deployments include:

• Adobe: Utilizing Foundry to enhance customer engagement and streamline creative workflows.
• Telefónica: Deploying agents to improve customer service operations and network management.
• Tata Consultancy Services (TCS): Leveraging Foundry to build and deploy AI solutions for enterprise clients across various industries, optimizing processes and driving innovation.
• Viva Republica (Toss): As mentioned, using the APAC Data Zone to ensure responsible data handling while accelerating AI innovation in the financial sector.

These examples underscore the platform’s ability to scale from concept to widespread adoption, proving its value in real-world business scenarios.

Getting Started with Microsoft Foundry

All the capabilities detailed in this announcement are now live and accessible within Microsoft Foundry.

To embark on your AI agent development journey, follow the comprehensive documentation available on Microsoft Learn. Developers can quickly get started by completing the quickstart guide, which provides a step-by-step walkthrough for setting up, testing, and deploying a production-ready hosted agent from end to end.

For a structured learning experience, the "AI Agents for Beginners" curriculum offers a 12-lesson series. Further deepening expertise can be achieved through guided labs such as "Develop AI Agents in Azure," the "Hosted Agents Workshop (.NET)," the "Foundry Toolkit for VS Code and hosted agents workshop," and the "ZavaShop Supply Chain Workshop." To ensure the quality and reliability of your agents, consult the practical guide on "Evaluating AI Agents: A Practical Guide with Microsoft Foundry."

For those seeking a visual and in-depth explanation, the "Foundry Agent Service + Microsoft Agent Framework Explained" video, featuring Jeff Hollan, provides a detailed walkthrough of operationalizing AI agents from deployment to achieving real-world impact.