Frontier fashions and manufacturing brokers: Advancing Microsoft Foundry for the agentic period

AI solely creates worth when it reveals up in actual programs—programs which can be dependable, observable, and aligned to enterprise outcomes. Greater than 100,000 organizations are already constructing on Microsoft Foundry, and corporations like Adobe, Telefónica, and Tata Consultancy Providers are working brokers in manufacturing at present.

At Microsoft Construct, we laid out a easy promise for the agentic period: builders ought to be capable of construct an agent the place they already work, run it on infrastructure they will belief, and put it in entrance of the individuals who want it— with out stitching collectively disconnected platforms. As we speak, that imaginative and prescient strikes from roadmap to actuality with three units of updates now typically obtainable in Microsoft Foundry:

• OpenAI’s newest frontier mannequin sequence: GPT-5.6 Sol, GPT-5.6 Terra, and GPT-5.6—every tuned to a special workload, obtainable in Commonplace World and Commonplace Information Zones.
• Asia-Pacific Information Zone, giving APAC clients a regional choice to run frontier OpenAI fashions whereas holding knowledge processing throughout the area.
• Manufacturing brokers in Foundry Agent Service, with hosted brokers, toolboxes, and publishing to Microsoft 365 Copilot and Microsoft Groups.

Collectively, these capabilities deliver frontier fashions, manufacturing agent runtime, enterprise-grade id, safety, and compliance controls, and distribution throughout Microsoft 365 right into a single platform—serving to organizations transfer from experimentation to manufacturing with out assembling disconnected instruments and companies.

Why Foundry is the perfect agent platform

Microsoft Foundry is Microsoft’s end-to-end platform for constructing, working, governing, and distributing AI brokers. Foundry brings collectively the capabilities organizations want to maneuver brokers into manufacturing throughout three pillars:

• Construct: Open and versatile throughout fashions and frameworks.
• Generate: Linked to enterprise knowledge, instruments, and customers.
• Govern: Secured, managed, and optimized for long-term worth.

These pillars come collectively within the newest Foundry updates, serving to organizations construct, run, and scale manufacturing brokers on a single platform.

Construct with any framework and mannequin on the {industry}’s end-to-end AI platform

Agent improvement begins the place builders already work—in GitHub Copilot and Microsoft Visible Studio (VS) Code—with the Foundry Toolkit for VS Code and the Foundry talent dealing with deployment to Foundry. Whether or not groups construct with Microsoft Agent Framework, GitHub Copilot SDK (typically obtainable), or Claude Agent SDK, Foundry is the manufacturing vacation spot—and all of it begins with the precise mannequin.

Begin with the precise mannequin for the precise job

An agent is just as succesful because the mannequin reasoning behind it. Microsoft Foundry provides organizations entry to industry-leading frontier, open-source, and task-specific fashions by a single platform, permitting groups to decide on the precise mannequin for each workload.

As we speak, we’re making OpenAI’s GPT-5.6 sequence typically obtainable in Microsoft Foundry Fashions and Microsoft Foundry Agent Service:

• GPT-5.6 Sol delivers essentially the most superior reasoning capabilities but, supporting prolonged reasoning, agentic workflows, and code-focused situations, for essentially the most demanding enterprise workloads.
• GPT-5.6 Terra is a balanced mannequin for on a regular basis work, delivering efficiency aggressive with GPT-5.5 at a decrease price, making it ultimate for scaling clever functions throughout the enterprise.
• GPT-5.6 Luna is the quickest and most inexpensive mannequin within the household, making it properly suited to high-volume, latency-sensitive workloads.

Collectively, the GPT-5.6 sequence provides organizations the flexibleness to match mannequin functionality, price, and efficiency necessities to particular enterprise situations, moderately than forcing each workload onto a single mannequin.

Clients persistently inform us that entry to new fashions issues as a lot as mannequin high quality. That’s why we’re making GPT-5.6 obtainable by World Commonplace and World Precedence Processing for all the present 28 world areas, Information Zones Commonplace, and World Provisioned from day one. This permits clients to undertake the newest frontier AI improvements the place they’ve already constructed, deployed, and scaled their functions.

GPT 5.6 pricing

Run frontier AI the place your corporation operates

Extra fashions, in additional areas, are solely half of what platform growth means. The opposite half is extra locations to run them compliantly. That is precisely what the Asia-Pacific Information Zone delivers. As we speak we’re additionally saying the common availability of the Asia-Pacific (APAC) Information Zone for Microsoft Foundry, enabling APAC clients to run frontier OpenAI fashions whereas holding knowledge processing throughout the Asia-Pacific areas, with no separate atmosphere to sew collectively and no ready for functionality to catch up.

With World, Information Zone, and Regional deployment choices obtainable in Foundry, organizations can align AI adoption with their sovereignty, compliance, efficiency, and scale necessities whereas sustaining a constant improvement and operations expertise throughout environments.

As monetary establishments undertake AI, accountable knowledge dealing with turns into foundational to belief. Microsoft Foundry’s APAC Information Zone permits us to maintain knowledge processing regionally anchored whereas accessing superior AI fashions at scale. This offers us the arrogance to speed up AI innovation responsibly and reinforces our ambition to be a number one AI-powered monetary platform in Asia.

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

Generate influence with action-oriented, context conscious brokers

A succesful mannequin is just the place to begin. To place one to work in manufacturing, an agent additionally wants someplace to run, data of your corporation, ruled entry to its instruments, reminiscence that carries throughout interactions and acts on real-world occasions, and a path to the folks who use it. Foundry offers every of these as a built-in functionality, designed to work collectively.

• The place it lives: Hosted brokers in Foundry Agent Service is now typically obtainable, giving builders one manufacturing runtime for brokers constructed with any framework and harness—Microsoft Agent Framework, GitHub Copilot SDK, LangGraph, OpenClaw, Hermes, and others. It’s enterprise-ready on day one: Community isolation with Microsoft Azure Digital Community (VNet) integration retains agent site visitors inside your safety boundary. For long-running workloads, the brand new resilient activity assist in hosted brokers (personal preview) makes it simpler to construct brokers that survive failures. The platform offers primitives to maintain the sandbox working, your harness offers checkpointing, and collectively they permit an agent to renew when it restarts. The outcome: multi-turn conversations, reasoning loops, and human-in-the-loop approvals can decide up the place they left off, with out builders having to construct their very own restoration, retry, and state-management.
• The way it talks: Hosted brokers with Voice Stay is now typically obtainable. Builders can add real-time voice experiences to the brokers they constructed with the frameworks they like, utilizing the Azure VoiceLive SDK. 
• What it is aware of: Foundry provides brokers entry to enterprise data with out requiring builders to construct a advanced retrieval pipeline from scratch. Microsoft IQ brings collectively Work IQ for real-time consciousness of your Microsoft 365 atmosphere, Material IQ in your structured knowledge, and Internet IQ for low-latency stay internet grounding—all unified behind Foundry IQ, now typically obtainable because the SLA-backed data layer behind each Foundry agent. 
• The way it reaches its instruments: Toolboxes in Foundry is mostly obtainable. As an alternative of transport each instrument definition on each request, a toolbox dynamically selects the precise instrument for the job—giving brokers ruled, curated entry whereas dramatically lowering the token overhead of enormous instrument units. 
• The way it remembers and responds to the world: Reminiscence and routines in Foundry Agent Service are in public preview. Reminiscence (procedural, person, and session) lets brokers carry context throughout interactions. Routines run any agent on a recurring schedule or timer so it acts with no person immediate; new event-based triggers, powered by the connector gateway, now let the identical agent get up the second an upstream system alerts a change—a ticket filed, a file touchdown in storage, a workflow finishing. 
• The way it reaches customers: Publishing to Microsoft Groups and Microsoft 365 Copilot is mostly obtainable subsequent week. The brokers your builders construct land within the functions the place tons of of thousands and thousands of individuals already do their work with id, permissions, and coverage flowing by robotically. And it really works even for network-isolated brokers: when a challenge runs behind a non-public endpoint, you publish by a documented stream moderately than the one-click button. The agent stays in your personal community whereas Microsoft’s channel adapters attain it by your personal firewall and reverse proxy.

Govern and optimize the total AI lifecycle with observability and controls

An agent you can’t see, enhance, or safe is an agent you can’t put in manufacturing—so Foundry treats belief as a platform precedence, not a developer duty. What’s new on this launch closes the loop round the whole lot that occurs after you construct: seeing what an agent did, making it higher, and proving it’s price working.

• The way you see what it did—tracing and analysis for hosted brokers, typically obtainable. See precisely what an agent did, why, and the place it went improper, and consider conduct systematically earlier than and after you ship.
• The way it will get higher and cheaper—agent optimizer in Foundry Agent Service, in public preview. It assessments your prompts, abilities, fashions, and instruments collectively and robotically identifies higher configurations—typically letting you maintain high quality whereas transferring to a smaller, cheaper mannequin.
• The way you show its worth—ROI for brokers in Microsoft Foundry, in personal preview. It connects an agent’s traces, business-value evaluations, and working price right into a single view—surfacing KPIs like web worth, whole price, and present ROI within the dashboard, so clients can see whether or not a manufacturing agent is creating extra worth than it prices to run, and drill into traces when it isn’t. 

As brokers scale from pilots to 1000’s of runs a day, Foundry provides groups the levers to maintain spend predictable with out leaving the platform.

That begins with alternative.

• Select the way you deploy. Foundry gives World, Information Zone, and Regional deployments, so you may align AI to your sovereignty, compliance, and efficiency necessities, working frontier fashions whereas holding knowledge processing in-region.
• Select the way you pay for mannequin inference. A full spectrum of gives—Commonplace, Precedence Processing, Provisioned Throughput, and Batch—permits you to optimize for agility, latency, throughput, and price on one platform.

On high of that basis, mannequin router matches every request to the precise mannequin, immediate caching cuts redundant computation, and PTU spillover and quota optimization protect service continuity by utilization spikes. For brokers, toolboxes in Foundry ship solely the instruments every request wants, and agent optimizer tunes prompts, abilities, instruments, and mannequin alternative in opposition to your personal evaluators.

And spend is just half the equation. ROI for brokers in Foundry connects enterprise worth, utilization, and price in a single view—so groups can see whether or not a manufacturing agent is creating extra worth than it prices to run, and the place price is outpacing worth.

For a hands-on walkthrough, watch our new Microsoft Mechanics episode on token economics for brokers.

In manufacturing: What groups are constructing on Foundry  

The organizations constructing on Foundry aren’t experimenting; they’re transport, from digital natives to the world’s largest enterprises.

• Adobe is constructing with GitHub, Foundry Agent Service, and Azure Features, deploying brokers for his or her functions and saving time and work attending to manufacturing.
• Telefónica has adopted Microsoft Foundry because the core of their company agentic platform, with the primary wave of brokers tackling community operations—a telco’s most advanced, strategic area—throughout Microsoft Agent Framework, hosted brokers, AI Gateway, and Azure Logic Apps.
• Tata Consultancy Providers is utilizing agent optimizer in Foundry Agent Service to enhance agent efficiency with a extra structured strategy to immediate tuning, serving to groups cut back handbook effort whereas measuring positive factors in activity adherence and execution effectivity.

The sample is constant: groups that when spent weeks integrating, securing, and deploying brokers are actually doing it in days, on infrastructure that meets their compliance bar, reaching customers by instruments they already belief.

Get began

Every little thing on this put up is stay in Microsoft Foundry.

Comply with the documentation and Microsoft Be taught programs. Builders can get began in minutes by following the Quickstart, which walks by organising, testing, and deploying a production-ready hosted agent finish to finish.

Try AI Brokers for Rookies for a 12-lesson curriculum, then go deeper with guided labs: Develop AI Brokers in Azure, Hosted Brokers Workshop (.NET), the Foundry Toolkit for VS Code and hosted brokers workshop, and the ZavaShop Provide Chain Workshop. To place your brokers on a stable high quality footing, learn Evaluating AI Brokers: A Sensible Information with Microsoft Foundry. 

Watch: Foundry Agent Service + Microsoft Agent Framework Defined—Jeff Hollan walks by methods to operationalize AI brokers from deployment to real-world influence.