📢 What's new in Azure AI Foundry | June 2025

[nicholasdbrady]

Welcome to the June 2025 edition of “What’s New in Azure AI Foundry!” This month brings major improvements and new features to help you build, deploy, and scale generative AI solutions more efficiently. Here’s a roundup of what’s arrived in the platform, including new models, agent tools, developer resources, and game-changing platform updates.

🚀 TL;DR – Key Highlights

• Agent Service is now GA with full Model Context Protocol (MCP) support – connect backend services using simple JSON-RPC without custom OpenAPI specs.
• Model launches:
  • o3-pro: Enterprise reasoning model for high-depth, consistent answers.
  • Sora: Public preview for text-to-video generation.
  • DeepSeek-R1: Advanced, open-source reasoning model.
  • Grok 3 & Grok 3 Mini: Real-time conversational and reasoning AI from xAI.
  • codex-mini: Lightweight coding assistant for efficient embedded scenarios.
• Platform Upgrades:
  • Unified Azure AI Foundry resource type with self-service Projects.
  • Easier API versioning: use latest or preview endpoints.
  • Unified SDK combining model, agent, and cloud integrations.
  • VS Code extension with full agent CRUD, IntelliSense, and deployment support.
• APIs & Tools:
  • Voice Live API (beta) for real-time speech.
  • Enhanced Responses API with MCP and tool-calling.
  • New integrations for Bing Custom Search, SharePoint, and Microsoft Fabric.
  • Model Safety Leaderboards now included in model benchmarks for responsible AI.

👉 Microsoft recognized as a Gartner Leader—second year in a row!
Learn how to scale AI transformation with proven frameworks.

---

🎉 Join the Azure AI Foundry Community

You’re never alone in your AI journey!

• 25,000+ developers are connecting on Discord and GitHub Discussions.
• Attend live events, AMAs, and weekly office hours.
• Access open-source courses, code samples, and learning paths for AI Agents, MCP, and GenAI.

👉 Join the Community

---

✨ Model Releases

• o3-pro: Enterprise Reasoning
  Highest reasoning performance for enterprise-grade, multimodal tasks.
• Sora: Video Generation (Preview)
  Generate 1080p videos from text prompts—text-to-video and (soon) image-to-video!
• codex-mini: Lightweight Code Assistant
  Efficient, cost-effective code generation for embedded scenarios.
• Grok 3 & Grok 3 Mini
  New conversational and reasoning models with competitive performance.
• DeepSeek-R1-0528
  Open-source, SOTA reasoning with chain-of-thought and reinforcement learning.

---

🧑‍💻 Agents & Tools

• MCP Agent Support (Preview): Use any MCP-compliant agent–no more custom backend code!
• SharePoint Tool Integration (Preview): Chat with private SharePoint docs using proper enterprise security.
• Multi-Agent Workflows: Orchestrate multiple agents in coordinated tasks.
• Unified Semantic Kernel + AutoGen: The new unified agent framework for all your composable AI needs.
• Catalog & Samples, Healthcare Tools: Prebuilt agent workflows, industry-focused orchestration.

---

🔧 Developer Experience & Platform Improvements

• VS Code Extension: Full agent CRUD, YAML IntelliSense, and more—without leaving your IDE.
• Trace Agents Debugging: Transparent insight into agent runs.
• GitHub Integration: Manage and deploy agents directly from your repository!
• Logic Apps Agent Triggers: Automate agent invocations for real-time business processes.
• Specialized tools: Bing Custom Search, Morningstar Research, improved Azure AI Search, and fine-tuning for o4-mini.

---

🛠️ Platform & API

• New unified resource type & self-service Projects
• Simplified API versions: latest / preview
• Unified SDK for models, agents & more
• Global fine-tuning
• Voice Live API (Beta)
• Model Safety Leaderboards
• Enhanced Responses API

---

🏁 Happy building!

Share your feedback and projects with #AzureAIFoundry & let us know what you create!

[michelschep]

A2A is not mentioned in the release notes.
It is however mentioned in some Microsoft blogposts.
Is that already a feature or coming soon in AI Foundry?

[amynic]

Hi @michelschep - check out this discussion about A2A in Agent Service (not yet released) but samples and workaround available https://github.com/orgs/azure-ai-foundry/discussions/92