Microsoft and xAI are thrilled to unveil the availability of Grok 3 and Grok 3 Mini into the Azure AI Foundry Models, marking a significant milestone in AI accessibility and innovation. This collaboration combines xAI’s cutting-edge models with Azure’s enterprise-ready infrastructure, giving developers access to Grok 3’s advanced capabilities in a secure, scalable environment.  Grok models enable a range of enterprise scenarios with advanced capabilities in reasoning, coding, and visual processing. 

Starting today, xAI’s flagship Grok 3 models are available for a free preview in Azure Foundry for the next 2 weeks. Grok models are also available to try on GitHub models.

Expanding the AI Ecosystem

Azure AI Foundry Models is a hub that hosts a wide range of models from Microsoft, partners and community. In fact, the model catalog features hundreds of foundation models across various providers – from OpenAI and Meta to Cohere, NVIDIA, Hugging Face, and more. The addition of xAI’s Grok 3 underscores Microsoft’s commitment to support an open, diverse AI ecosystem, rather than relying on a single model provider. Grok 3’s arrival on Azure AI Foundry Models is a testament to that vision, bringing a fresh new model into the fold and expanding the toolkit available to developers.

What is Grok?

Grok, developed by xAI, is a family of large language models designed to push the boundaries of AI innovation and accelerate human scientific discovery. According to xAI, Grok 3, the flagship model, blends unparalleled reasoning with vast pretraining knowledge, honed on xAI’s Colossus supercluster with 10x the compute power of prior leading models. Grok 3 models excel in reasoning, mathematics, coding, world knowledge, and instruction-following, setting a new standard for enterprise AI. Grok 3 Mini, a lightweight yet powerful thinking model, is optimized for logic-based tasks, delivering efficiency without compromise.

Key features of Grok 3 and Grok 3 Mini:

• Advanced reasoning with reinforcement learning: Grok 3 Mini’s advanced reasoning capabilities, powered by large-scale reinforcement learning (RL) enable it to tackle complex problems with great accuracy. Grok 3 Mini can backtrack, correct errors, and explore multiple solution paths—delivering transparent, actionable insights for enterprises through its chain-of-thought reasoning.
• Deep domain expertise: With deep domain expertise in finance, healthcare, law and science, Grok 3 excels at enterprise tasks like financial forecasting, medical diagnosis support, legal document analysis, and scientific research assistance—delivering precise, domain-specific solutions.
• Extended Context Length: With an extended context length of up to 131K tokens, Grok processes and understands vast datasets in a single pass—ideal for comprehensive analysis of large documents or complex workflows.
• Reasoning effort parameter: For more fine-grained control over the model’s performance, Grok 3 Mini supports the reasoning effort parameter, which allows users to adjust the model’s thinking effort with options for low and high reasoning levels.
• Structured outputs: xAI’s Grok 3 models support structured outputs, enabling developers to specify JSON schemas for AI-powered automations.
• Functions and Tools support: Like other Azure AI Foundry Models, Grok 3 models support functions and external tools that enable enterprises to build agentic workflows. 

Performance

According to X.ai, Grok 3 and Grok 3 Mini have demonstrated exceptional results on several academic and real-world benchmarks: 

Grok 3:  

• 60% on AIME 2025 (math competition)
• 79.1% on GPQA (graduate-level reasoning)
• 65.5% on LiveCodeBench (code generation)
• 91.1% on IFEval (instruction following)
• 77.4% on TauBench-Retail (agentic shopping)
• 43.0% on TauBench-Airline (agentic flight booking)
• 83.1% on MMLU-pro (multi-task language understanding)

Grok 3 Mini:  

• 90.7% on AIME 2025 (math competition)
• 80.3% on GPQA (graduate-level reasoning)
• 74.8% on LiveCodeBench (code generation)
• 82.8% on MMLU-pro (multi-task language understanding)

Seamless Access to Grok 3 via Azure AI Foundry

It is easy for developers to start using Grok 3 on Azure AI Foundry Models where you can easily discover, deploy, and manage AI models. Grok 3 now appears in the catalog alongside other foundation models. Just as you would with an OpenAI or Hugging Face model, you can quickly find Grok 3’s model card (with details and benchmarks) and deploy it in a few clicks 

For integration into applications, flexible deployment options are supported for models available direct from Microsoft in Azure AI Foundry:

• Standard (Pay-Go)– You can call Grok 3 as a managed API endpoint (pay-per-token). No infrastructure to manage: just send your prompts to the Azure endpoint and get responses. This is great for getting started quickly or for scaling on-demand without upfront provisioning.
• Azure AI Foundry Provisioned Throughput (PTUs) – Grok 3 will soon support PTU deployments on Azure. This means you can reserve a fixed throughput capacity for the model, allowing you access to dedicated resources to handle a certain rate of requests. Provisioned throughput deployments provide predictable latency and performance by allocating processing power in advance. For production scenarios where you expect steady high volume or need strict latency, provisioning Grok 3 with PTUs can be cost-effective and reliable.

Pricing

Starting May 19, Grok 3 and Grok 3 Mini models will be available at no cost in Azure AI Foundry Models for a limited free preview. Customers can explore and test both models free of charge through early June. 

Beginning in June, pricing as shown in table below will apply for Grok 3 and Grok 3 Mini deployments in Azure AI Foundry.

Enterprise-Grade Features: Scale, Safety, and Observability

When you use Grok 3 through Azure AI Foundry, you’re not just getting raw model access – you’re getting a whole suite of tools and services to support your AI application’s lifecycle. Running Grok 3 on Azure comes with the same level of enterprise readiness that developers expect. Here are a few key benefits:

• Scalable Deployments with PTUs: Coming soon in early June you will be able to deploy Grok 3 with provisioned throughput units for mission-critical workloads. This allows you to have dedicated processing capacity  and stable performance even under heavy load. You can scale up or down the PTU allocation based on your needs, and Azure will handle the backend hardware management. For developers, this means less time worrying about infrastructure and more time building features. If your application usage is sporadic or in early stages, you can start with the standard serverless mode (pay-go) and later upgrade to PTUs once you have predictable traffic. Azure’s flexibility here supports both experimentation and production scaling.
• Built-in Content Safety Controls: Azure provides granular content safety tools to keep outputs from Grok 3 models and other Azure AI Foundry Models compliant with your requirements. Azure AI Foundry integrates with the Azure AI Content Safety service, which can detect and filter content that violates predefined policies (for hate, violence, sexual content, etc. You have the option to enable content filtering on Grok 3’s responses via serverless inference APIs, or to route outputs through content moderation APIs for managed deployments. Essentially, you can dial the safety settings up or down to match your use case – whether you want maximum openness for research or strict guardrails for a customer-facing app. All content filtering and policy enforcement is handled within Azure’s trusted framework, so you can confidently deploy Grok 3 in sensitive or regulated environments.
• Observability and Monitoring: Azure AI Foundry offers observability for all deployed models, including Grok 3. Out of the box, you can monitor usage metrics like token consumption, request rates, and latency through Azure Monitor. Even more, Azure AI Foundry integrates with Azure Monitor Application Insights to provide deep telemetry and tracing of your AI calls.  Developers can instrument their applications with the Azure AI tracing SDK (which supports OpenTelemetry standards) to log detailed traces of each request/response, and then visualize this data in Application Insights dashboards. This means you can track how Grok 3 is performing in real time – spotting errors, measuring response quality, and even setting up alerts (for example, to notify you if response time exceeds a threshold or if content safety flags a high-severity issue). Continuous evaluation tools are also available: Azure AI Foundry lets you define evaluation metrics and run them against the model’s outputs continuously. In short, you get a full GenAI Ops toolkit so that your Grok 3-powered application is reliable, safe, and high-quality over time.

By combining Grok 3’s raw capabilities with Azure’s enterprise features, developers can confidently move from prototype to production. You get the best of both worlds: a model that pushes the envelope of what AI can do, and a platform that mitigates the risks and challenges of using such a model in real apps. 

Get Started with Grok 3 Today

The launch of Grok 3 on Azure AI Foundry Models is an exciting development for the AI developer community. It combines the bold innovation of xAI’s model with the robust tools of Azure’s platform. Whether you’re building intelligent assistants, processing large documents, or exploring new AI frontiers, Grok 3 gives you a powerful new option to consider. Head over to Azure AI Foundry and give Grok 3 a try – we can’t wait to see what you build with it! Get started with Azure AI Foundry.