Transitioning from Azure Language Features to Foundry Models

2026-06-10 · Source: Microsoft Foundry Blog articles · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Cloud Computing & IT Infrastructure, Software Development & Engineering · Depth: Intermediate, extended

Summary

Eight Azure AI Language features, including Key Phrase Extraction, Sentiment Analysis, Summarization, and Custom Text Classification, are being deprecated, with final retirement by March 2029 (Entity Linking by September 20, 2028). Microsoft recommends transitioning these functionalities to Foundry-based models, utilizing a unified inference endpoint. This transition involves replacing existing features with solutions built on foundation models like Mistral-small-2503, Phi-3.5-mini, Claude Sonnet 4.6, and GPT-4o, often combined with prompt engineering, retrieval-augmented generation (RAG), or fine-tuning. Microsoft's new MAI-Thinking-1 model is also available in private preview for reasoning-heavy tasks. The guidance provides step-by-step tutorials to implement these new solutions via Foundry's REST API, ensuring structurally similar JSON outputs for continuity.

Key takeaway

For AI Engineers and Software Engineers currently relying on deprecated Azure AI Language features, you must proactively transition your applications to Microsoft Foundry models. Begin by identifying affected services and re-architecting them using recommended foundation models, prompt engineering, or fine-tuning strategies. This ensures continuity of service and upgrades capabilities, moving from static APIs to flexible, continuously improving LLMs on a unified platform. Plan your migration now to avoid service disruption before the March 2029 retirement deadline.

Key insights

Azure AI Language features are deprecating, requiring migration to Foundry models using LLMs, RAG, or fine-tuning for equivalent functionality.

Principles

• Foundation models can replicate specialized AI services.
• Prompt engineering is crucial for structured LLM outputs.
• Fine-tuning provides domain-specific accuracy and control.

Method

Transition involves deploying a Foundry model, crafting a task-specific prompt, calling the unified REST API, and processing the structured JSON output.

In practice

• Replicate Key Phrase Extraction using mistral-small-2503 with JSON prompts.
• Implement Entity Linking via gpt-4o and Azure AI Search for knowledge retrieval.
• Fine-tune Ministral-3B for custom text classification on domain-specific data.

Topics

• Azure AI Language
• Microsoft Foundry
• Large Language Models
• Prompt Engineering
• Model Fine-tuning
• Retrieval-Augmented Generation

Best for: AI Engineer, Machine Learning Engineer, Software Engineer

Related on AIssential

• New Azure Open AI models bring fast, expressive, and real‑time AI experiences in Microsoft Foundry — New Azure OpenAI models prioritize continuity and reliability for complex, real-time AI applications and long-running engineering tasks.
• [NOTICE] This feed is no longer maintained — The Mastering LLMs conference provides practical, unbiased, and diverse expert insights into LLM deployment and fine-tuning.
• [AINews] The End of Finetuning — Finetuning is declining for general AI engineering but remains critical for top-tier applications and open models.
• Moving beyond manual prompting: A practical introduction to DSPy — DSPy offers a programmatic, declarative framework for building robust, optimizable LLM applications, moving beyond manual prompt engineering.
• Large Language Models in NLP — LLMs integrate NLP advancements with massive scale to create general-purpose, prompt-driven language systems.
• The hard problems were never language problems. — Production AI agents require robust system architectures beyond just Large Language Models.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Microsoft Foundry Blog articles.