With the launch of OpenAI’s gpt‑oss fashions—its first open-weight launch since GPT‑2—we’re giving builders and enterprises unprecedented capacity to run, adapt, and deploy OpenAI fashions totally on their very own phrases. For the primary time, you possibly can run OpenAI fashions like gpt‑oss‑120b on a single enterprise GPU—or run gpt‑oss‑20b regionally.
AI is not a layer within the stack—it’s changing into the stack. This new period requires instruments which might be open, adaptable, and able to run wherever your concepts stay—from cloud to edge, from first experiment to scaled deployment. At Microsoft, we’re constructing a full-stack AI app and agent manufacturing unit that empowers each developer not simply to make use of AI, however to create with it.
That’s the imaginative and prescient behind our AI platform spanning cloud to edge. Azure AI Foundry gives a unified platform for constructing, fine-tuning, and deploying clever brokers with confidence whereas Foundry Native brings open-source fashions to the sting—enabling versatile, on-device inferencing throughout billions of gadgets. Home windows AI Foundry builds on this basis, integrating Foundry Native into Home windows 11 to help a safe, low-latency native AI improvement lifecycle deeply aligned with the Home windows platform.
With the launch of OpenAI’s gpt‑oss fashions—its first open-weight launch since GPT‑2—we’re giving builders and enterprises unprecedented capacity to run, adapt, and deploy OpenAI fashions totally on their very own phrases.
For the primary time, you possibly can run OpenAI fashions like gpt‑oss‑120b on a single enterprise GPU—or run gpt‑oss‑20b regionally. It’s notable that these aren’t stripped-down replicas—they’re quick, succesful, and designed with real-world deployment in thoughts: reasoning at scale within the cloud, or agentic duties on the edge.
And since they’re open-weight, these fashions are additionally simple to fine-tune, distill, and optimize. Whether or not you’re adapting for a domain-specific copilot, compressing for offline inference, or prototyping regionally earlier than scaling in manufacturing, Azure AI Foundry and Foundry Native provide the tooling to do all of it—securely, effectively, and with out compromise.
Open fashions, actual momentum
Open fashions have moved from the margins to the mainstream. Right this moment, they’re powering every thing from autonomous brokers to domain-specific copilots—and redefining how AI will get constructed and deployed. And with Azure AI Foundry, we’re providing you with the infrastructure to maneuver with that momentum:
- With open weights groups can fine-tune utilizing parameter-efficient strategies (LoRA, QLoRA, PEFT), splice in proprietary information, and ship new checkpoints in hours—not weeks.
- You possibly can distill or quantize fashions, trim context size, or apply structured sparsity to hit strict reminiscence envelopes for edge GPUs and even high-end laptops.
- Full weight entry additionally means you possibly can examine consideration patterns for safety audits, inject area adapters, retrain particular layers, or export to ONNX/Triton for containerized inference on Azure Kubernetes Service (AKS) or Foundry Native.
Briefly, open fashions aren’t simply feature-parity replacements—they’re programmable substrates. And Azure AI Foundry gives coaching pipelines, weight administration, and low-latency serving backplane so you possibly can exploit each a kind of levers and push the envelope of AI customization.
Meet gpt‑oss: Two fashions, infinite prospects
Right this moment, gpt‑oss-120b and gpt‑oss-20b can be found on Azure AI Foundry. gpt‑oss-20b can also be accessible on Home windows AI Foundry and might be coming quickly on MacOS by way of Foundry Native. Whether or not you’re optimizing for sovereignty, efficiency, or portability, these fashions unlock a brand new stage of management.
- gpt‑oss-120b is a reasoning powerhouse. With 120 billion parameters and architectural sparsity, it delivers o4-mini stage efficiency at a fraction of the dimensions, excelling at complicated duties like math, code, and domain-specific Q&A—but it’s environment friendly sufficient to run on a single datacenter-class GPU. Perfect for safe, high-performance deployments the place latency or value matter.
- gpt‑oss-20b is tool-savvy and light-weight. Optimized for agentic duties like code execution and power use, it runs effectively on a variety of Home windows {hardware}, together with discrete GPUs with16GB+ VRAM, with help for extra gadgets coming quickly. It’s good for constructing autonomous assistants or embedding AI into real-world workflows, even in bandwidth-constrained environments.
Each fashions will quickly be API-compatible with the now ubiquitous responses API. Meaning you possibly can swap them into present apps with minimal modifications—and most flexibility.
Bringing gpt‑oss to Cloud and Edge
Azure AI Foundry is greater than a mannequin catalog—it’s a platform for AI builders. With greater than 11,000 fashions and rising, it offers builders a unified house to judge, fine-tune, and productionize fashions with enterprise-grade reliability and safety.
Right this moment, with gpt‑oss within the catalog, you possibly can:
- Spin up inference endpoints utilizing gpt‑oss within the cloud with only a few CLI instructions.
- Wonderful-tune and distill the fashions utilizing your personal information and deploy with confidence.
- Combine open and proprietary fashions to match task-specific wants.
For organizations growing eventualities solely doable on shopper gadgets, Foundry Native brings distinguished open-source fashions to Home windows AI Foundry, pre-optimized for inference by yourself {hardware}, supporting CPUs, GPUs, and NPUs, via a easy CLI, API, and SDK.
Whether or not you’re working in an offline setting, constructing in a safe community, or operating on the edge—Foundry Native and Home windows AI Foundry helps you to go absolutely cloud-optional. With the potential to deploy gpt‑oss-20b on fashionable high-performance Home windows PCs, your information stays the place you need it—and the facility of frontier-class fashions involves you.
That is hybrid AI in motion: the flexibility to combine and match fashions, optimize efficiency and price, and meet your information the place it lives.
Empowering builders and choice makers
The supply of gpt‑oss on Azure and Home windows unlocks highly effective new prospects for each builders and enterprise leaders.
For builders, open weights imply full transparency. Examine the mannequin, customise, fine-tune, and deploy by yourself phrases. With gpt‑oss, you possibly can construct with confidence, understanding precisely how your mannequin works and easy methods to enhance it in your use case.
For choice makers, it’s about management and adaptability. With gpt‑oss, you get aggressive efficiency—with no black packing containers, fewer trade-offs, and extra choices throughout deployment, compliance, and price.
A imaginative and prescient for the longer term: Open and accountable AI, collectively
The discharge of gpt‑oss and its integration into Azure and Home windows is a part of a much bigger story. We envision a future the place AI is ubiquitous—and we’re dedicated to being an open platform to deliver these revolutionary applied sciences to our clients, throughout all our information facilities and gadgets.
By providing gpt‑oss via a wide range of entry factors, we’re doubling down on our dedication to democratize AI. We acknowledge that our clients will profit from a various portfolio of fashions—proprietary and open—and we’re right here to help whichever path unlocks worth for you. Whether or not you’re working with open-source fashions or proprietary ones, Foundry’s built-in security and safety instruments guarantee constant governance, compliance, and belief—so clients can innovate confidently throughout all mannequin sorts.
Lastly, our help of gpt-oss is simply the most recent in our dedication to open instruments and requirements. In June we introduced that GitHub Copilot Chat extension is now open supply on GitHub below the MIT license—step one to make VS Code an open supply AI editor. We search to speed up innovation with the open-source neighborhood and drive larger worth to our market main developer instruments. That is what it seems like when analysis, product, and platform come collectively. The very breakthroughs we’ve enabled with our cloud at OpenAI at the moment are open instruments that anybody can construct on—and Azure is the bridge that brings them to life.
Subsequent steps and assets for navigating gpt‑oss
- Deploy gpt‑oss within the cloud at this time with a couple of CLI instructions utilizing Azure AI Foundry. Browse the Azure AI Mannequin Catalog to spin up an endpoint.
- Deploy gpt‑oss-20b in your Home windows machine at this time (and shortly on MacOS) by way of Foundry Native. Comply with the QuickStart information to study extra.
- Pricing1 for these fashions is as follows:
*See Managed Compute pricing web page right here.
1Pricing is correct as of August 2025.