GPT-5.6 now available in Microsoft Foundry 

AI truly creates value when it’s integrated into practical systems that are dependable, trackable, and geared towards achieving business goals. Over 100,000 companies are currently utilising Microsoft Foundry, with big names like Adobe, Telefónica, and Tata Consultancy Services operating agents in real-world applications.

During Microsoft Build, we made a straightforward promise for the agentic era: developers ought to be able to create agents within their familiar environments, utilising trusted infrastructure, and easily providing access to users who need them—without the hassle of piecing together various unconnected platforms. Today, we’re delighted to announce that this vision has become a reality, thanks to three exciting updates that are now generally available in Microsoft Foundry:

• OpenAI’s latest frontier model series: Including GPT-5.6 Sol, GPT-5.6 Terra, and GPT-5.6—each optimally tuned for specific workloads, accessible in Standard Global and Standard Data Zones.
• Asia-Pacific Data Zone: This new option allows customers in the APAC region to run frontier OpenAI models while ensuring their data processing remains local.
• Production agents in Foundry Agent Service, featuring hosted agents, toolboxes, and the option to publish to Microsoft 365 Copilot and Microsoft Teams.

These innovations combine frontier models, a production agent runtime, enterprise-level identity, security, and compliance controls, along with distribution throughout Microsoft 365, all on a single platform. This greatly assists organisations as they transition from trial and error to actual production without needing to cobble together disparate tools and services.

Why Foundry is the best agent platform

Microsoft Foundry represents Microsoft’s all-in-one platform for crafting, deploying, overseeing, and distributing AI agents. Foundry consolidates everything organisations require to move their agents into production into three core areas:

• Build: Open and adaptable across models and frameworks.
• Generate: Seamlessly linked to enterprise data, tools, and users.
• Govern: Secured, effectively managed, and optimised for long-term success.

These essential elements come together with the latest Foundry updates, enabling organisations to build, run, and scale their production agents on a unified platform.

Build with any framework and model on the industry’s end-to-end AI platform

Developing agents begins in the tools developers already use—like GitHub Copilot and Microsoft Visual Studio (VS) Code. With the Foundry Toolkit for VS Code and the Foundry skill taking care of deployment to Foundry, teams can work efficiently. Whether they employ Microsoft Agent Framework, GitHub Copilot SDK (now widely available), or Claude Agent SDK, Foundry is the ultimate destination, with everything commencing from the right model choice.

Start with the right model for the right job

An agent’s effectiveness hinges on the model it operates on. Microsoft Foundry opens doors to industry-leading frontier, open-source, and task-specific models via a single platform, empowering teams to select the most suitable model for each workload.

We’re thrilled to announce that OpenAI’s GPT-5.6 series is now generally available in both Microsoft Foundry Models and Microsoft Foundry Agent Service:

• GPT-5.6 Sol: This model offers exceptional reasoning capabilities for rigorous enterprise workloads, supporting extended reasoning and agentic workflows.
• GPT-5.6 Terra: A balanced everyday model, it delivers performance on par with GPT-5.5 but at a reduced cost, ideal for scaling intelligent applications across organisations.
• GPT-5.6 Luna: The fastest and most cost-effective option, perfect for high-volume, latency-sensitive tasks.

The GPT-5.6 series grants businesses the flexibility to align model capability, cost, and performance with distinct business needs—rather than being forced to apply a one-size-fits-all approach to every task.

Our clients consistently express that having access to new models is just as important as the quality of the models themselves. Therefore, we’re rolling out GPT-5.6—the latest innovation—through Global Standard and Global Priority Processing for all 28 existing global regions, along with Data Zones Standard and Global Provisioned from day one. This allows clients to embrace cutting-edge frontier AI innovations within their already established applications.

GPT-5.6 pricing for Sol, Terra, and Luna

Below is the pricing table for GPT-5.6 models—Sol, Terra, and Luna—in Microsoft Foundry. Use this information to compare model options and estimate deployment expenses:

Run frontier AI where your business operates

Having access to more models in a wider range of regions is just part of what expanding the platform means. It’s equally important to ensure their compliant operation. This is precisely what the Asia-Pacific Data Zone offers. We are pleased to announce the general availability of the Asia-Pacific (APAC) Data Zone for Microsoft Foundry, allowing APAC clients to operate frontier OpenAI models while processing data locally, avoiding the complications of separate environments and waiting for capabilities.

With Global, Data Zone, and Regional deployment options within Foundry, businesses can align their AI initiatives with their sovereignty, compliance, performance, and scalability needs while maintaining a consistent development and operational experience across all environments.

As financial institutions increasingly adopt AI, responsible data handling is crucial for building trust. Microsoft Foundry’s APAC Data Zone allows us to manage data processing regionally while gaining access to advanced AI models at scale. This empowers us to accelerate AI innovation responsibly and reinforces our goal to lead as an AI-driven financial platform in Asia.

—Hongsoo Kim, Chief Data and AI Officer (CDAO), Viva Republica (Toss)

Generate impact with action-oriented, context-aware agents

A powerful model is merely the beginning. To successfully deploy it in production, agents must also have a platform to operate, knowledge of your business, controlled access to necessary tools, memory to recall previous interactions, and a direct line to the users. Foundry offers all these as built-in functionalities, specifically designed to work in harmony.

• Where it lives: Hosted agents in Foundry Agent Service are now readily available, providing developers a single production environment for agents developed in any framework—be it Microsoft Agent Framework, GitHub Copilot SDK, LangGraph, OpenClaw, Hermes, or more. It’s enterprise-ready right from the start, featuring network isolation with Microsoft Azure Virtual Network (VNet) integration that keeps agent traffic within your security perimeter. New resilient task support in hosted agents (currently in private preview) facilitates creating agents capable of surviving disruptions, allowing them to maintain multi-turn conversations and complete reasoning loops without requiring excessive developer intervention.
• How it communicates: With Hosted agents and Voice Live, developers can incorporate real-time voice interactions in their agents built on preferred frameworks using the Azure VoiceLive SDK.
• What it knows: Foundry simplifies access to enterprise knowledge for agents, eliminating the need for developers to create complex retrieval systems. Microsoft IQ unites Work IQ for real-time insights into your Microsoft 365 environment, Fabric IQ for structured data, and Web IQ for rapid real-time web grounding—all interlinked under Foundry IQ, which is now generally available as the knowledge layer underpinning every Foundry agent.
• How it accesses tools: Toolboxes in Foundry are now available. Instead of sending every tool definition with each request, a toolbox intelligently selects the best-suited tool, giving agents curated access while significantly reducing token overhead from large tool collections.
• How it recalls and responds to changes: The Memory and routines within Foundry Agent Service are now in public preview. Memory (including procedural, user, and session) allows agents to retain context over interactions. Routines enable any agent to run on a set schedule or whenever a designated event occurs, keeping them alert to changes, whether it’s a ticket submission, a new file, or the completion of a workflow.
• How it connects with users: Publishing to Microsoft Teams and Microsoft 365 Copilot will be generally available next week. The agents developed by your team will be integrated into applications that millions of users rely on daily, with identity, permissions, and policies seamlessly embedded. Even network-isolated agents can connect when run behind a private endpoint, utilizing a documented flow instead of the standard one-click button. This way, the agents remain within your private network while Microsoft’s channel adapters facilitate connection through your firewall.

Govern and optimise the entire AI lifecycle with observability and controls

An agent that lacks visibility, ongoing improvement, or security cannot be effectively put into production—which is why Foundry prioritises trust at the platform level rather than placing the entire burden on developers. The new features in this release provide comprehensive insights into everything that occurs after an agent is built: monitoring its actions, enhancing its capabilities, and justifying its role in your operations.

• Understanding agent actions: Tracing and evaluation for hosted agents are now available. You can see precisely what an agent did, why it did it, and where issues arose, while systematically evaluating its performance both before and after it goes live.
• Enhancing performance and reducing costs: Agent optimiser in Foundry Agent Service is in public preview. It tests your prompts, skills, models, and tools collectively, automatically identifying better configurations that often allow you to maintain quality while opting for smaller, more economical models.
• Demonstrating value: ROI for agents in Microsoft Foundry is in private preview. This feature consolidates an agent’s traces, evaluations of business value, and operating costs into a single view—highlighting key performance indicators such as net value, total costs, and current ROI on the dashboard, allowing teams to see if a production agent is providing greater value than its operational costs and delve into traces if it isn’t.

As agents transition from pilot tests to thousands of runs daily, Foundry provides the tools necessary to keep expenses predictable without requiring teams to switch platforms.

This approach starts with offering flexibility:

• Choose your deployment method: Foundry supports Global, Data Zone, and Regional deployments, enabling alignment with your sovereignty, compliance, and performance standards while running frontier models and maintaining local data processing.
• Decide how to pay for model inference: A broad spectrum of choices—Standard, Priority Processing, Provisioned Throughput, and Batch—allows optimisation according to agility, latency, throughput, and cost on one platform.

Beyond that, model router ensures that each request is matched to the appropriate model. Meanwhile, prompt caching minimises redundant computations, and PTU spillover and quota optimisation guarantee service continuity during usage spikes. For agents, toolboxes in Foundry only deploy the necessary tools for each request, and the agent optimiser adjusts prompts, skills, tools, and model choices based on your evaluations.

Furthermore, spending is just one aspect to consider. ROI for agents in Foundry links business value, usage, and cost into a unified view, enabling teams to monitor whether a production agent is generating more value than it incurs costs and identify areas where expenditures exceed value.

For a practical walkthrough, check out our new Microsoft Mechanics episode covering token economics for agents.

In production: What teams are building on Foundry

The organisations utilising Foundry are not merely experimenting; they are actively deploying solutions, from digital innovators to some of the largest companies worldwide.

• Adobe is developing with GitHub, Foundry Agent Service, and Azure Functions, rolling out agents for their applications and significantly reducing the time and effort required to reach production.
• Telefónica has integrated Microsoft Foundry as the backbone of their corporate agentic platform, with initial agents focusing on network operations—the one of the most intricate and strategic areas in telecommunications—leveraging Microsoft Agent Framework, hosted agents, AI Gateway, and Azure Logic Apps.
• Tata Consultancy Services employs the agent optimiser within Foundry Agent Service to enhance agent performance through a methodical approach to prompt tuning, which minimises manual effort while gauging improvements in task adherence and execution efficiency.

The trend is clear: teams that once dedicated weeks to integrating, securing, and deploying agents are now achieving this in mere days, thanks to infrastructure that meets their compliance demands and reaching users through familiar tools.

Get started

Everything discussed in this post is live within Microsoft Foundry.

For comprehensive guidance, follow the documentation and Microsoft Learn courses. Developers can kick off projects in minutes by following the Quickstart, which guides you through the process of setting up, testing, and deploying a production-ready hosted agent.

Explore AI Agents for Beginners for a structured 12-lesson course. Then dive deeper through guided labs: Develop AI Agents in Azure, Hosted Agents Workshop (.NET), the Foundry Toolkit for VS Code and hosted agents workshop, and the ZavaShop Supply Chain Workshop. For insights on establishing a solid quality framework for your agents, check out Evaluating AI Agents: A Practical Guide with Microsoft Foundry.

Watch: Foundry Agent Service + Microsoft Agent Framework Explained—Jeff Hollan walks you through how to operationalise AI agents from deployment to their real-world impact.

FAQs

• What is Microsoft Foundry?
  Microsoft Foundry is an end-to-end platform designed for building, running, managing, and distributing AI agents across various frameworks and models.
• How do I get started with Foundry?
  You can start by following the documentation available on Microsoft Foundry’s website, along with a Quickstart guide for setting up hosted agents.
• What models are available in Foundry?
  Foundry provides access to the GPT-5.6 model series, including Sol, Terra, and Luna, which are tailored for different use cases and workloads.
• What are the pricing options for GPT-5.6?
  Pricing varies by model, with GPT-5.6 Sol, Terra, and Luna having distinct costs based on input and output tokens. Please refer to the pricing table for details.
• Can I publish agents to Microsoft Teams?
  Yes, you can publish your agents to Microsoft Teams and Microsoft 365 Copilot, making them accessible to a vast user base seamlessly.