Microsoft’s July 2025 Azure AI Foundry release delivers significant upgrades to models, agents, developer tools, and platform capabilities, making it easier to build, monitor, and scale AI solutions securely.
Key Highlights:
Models
- GPT-image-1 now supports input fidelity for preserving style and features during image edits, and partial image streaming for early visual previews during generation.
Agents
- Deep Research Agent (Public Preview) enables automated, multi-step web research using OpenAI’s o3-deep-research model with Bing Search for authoritative results.
Tools
- AgentOps adds advanced tracing, evaluation, and monitoring for agents; AI Red Teaming Agent (Preview) now offers a cloud-based UI for reviewing security and compliance tests.
SDK Updates
- Major updates to Python, .NET, Java, and JavaScript/TypeScript SDKs, including new agent features, improved evaluation tools, and some breaking changes in method names and APIs.
Fine-tuning & Evaluation
- New public previews: RFT Observability (real-time monitoring of fine-tuning jobs), Quick Evals (fast output checks), Python Grader (custom evaluation logic). Developer Tier is now GA for 24-hour free model hosting to aid prototyping.
Platform & API Enhancements
- MCP connections in public preview, Model Router and Leaderboard available via SDKs/APIs, reinforcement fine-tuning support for select models.
Documentation & Guidance
- New Terraform deployment guide, Deep Research samples, updated model availability, new PII filtering docs, fine-tuning cost management guide, and increased file upload limits (300GB).