As organizations increasingly tailor foundation models to meet their domain-specific needs, Azure AI Foundry continues to deliver new capabilities that streamline, scale, and enhance the fine-tuning experience. One such organization, Decagon AI, fine-tuned GPT-4o-mini using Azure OpenAI Service’s supervised fine-tuning for their customer service agents. They were able to improve model accuracy and observed substantially lower latency for inferencing. This is one of my favorite use cases because it combines two cutting edge techniques – agents and fine tuning – for better results!

“Fine-tuning GPT-4o-mini on Azure dramatically accelerated our delivery timeline,” said Ashwin Sreenvias, CEO at Decagon AI. “The training performance, simplicity of the pipeline, and integrated tooling gave us a huge edge.”

This April, we introduced a series of updates to further empower developers and ML practitioners with more control, flexibility, and efficiency across their fine-tuning workflows.

What’s new?

Fine-tuning support for GPT-4.1 and GPT-4.1-mini 

We’re excited to announce fine-tuning support for the latest GPT-4.1 and GPT-4.1-mini models in Azure OpenAI Service. These models offer significant improvements in reasoning and instruction following, making them ideal for enterprise use cases that demand high accuracy and nuanced understanding. Now, customers can fine-tune these models using supervised techniques to reflect their organization’s tone of voice, internal workflows, and domain-specific knowledge. This enables more precise, aligned, and context-aware outputs across applications like customer support, financial advisory, document summarization, and internal copilots. 

GPT-4.1 and GPT-4.1-mini are available for training on North Central US and Sweden Central with Global Standard tier deployment. For the latest model region updates please check here. 

Fine-tuning support for third-party models, now in more regions

Building on our previous release of serverless fine-tuning for third-party models, we are expanding regional availability to include additional U.S. regions. This ensures lower latency and better compliance with data residency requirements for organizations fine-tuning models such as Phi-4-mini-instruct and Mistral-Large-2411. With serverless infrastructure managed by Azure AI Foundry, teams can launch fine-tuning jobs without provisioning compute resources, enabling faster time to value and effortless scalability. Whether you’re customizing open models for knowledge-intensive chatbots or streamlining domain-specific tasks, Azure AI Foundry offers a unified and flexible experience. 

Introducing the Evaluation API: Code-first grading at scale

To complement the recent release of the Stored Completions API, we’re launching the new Evaluation API—a powerful tool for assessing model output quality with automation and precision. This API allows developers to score completions using built-in graders (e.g., string match, object comparison) or define custom graders tailored to their use cases. It simplifies A/B testing, regression validation, and iterative model refinement by enabling scalable, programmatic evaluation loops. Paired with Stored Completions, developers can now generate, store, grade, and fine-tune model responses entirely in code, reducing manual overhead and accelerating the distillation cycle. Read more about it here. 

Stay tuned for more! 

These updates reflect our continued investment in making fine-tuning more accessible, reliable, and scalable across use cases and model families. Whether you’re optimizing for accuracy, speed, tone, or efficiency, Azure AI provides the tools you need to adapt frontier models to your real-world scenarios. 

Stay tuned for more updates as we expand support for newer models, techniques like reinforcement fine-tuning, and integrated agent-level tuning workflows in Azure AI Foundry. 

Resources 

• Explore the documentation to learn more about fine-tuning in Azure AI Foundry  

• Attend my session at Microsoft Build to learn about the latest updates on fine-tuning and distillation in Azure AI Foundry. 

• Get started today with Azure OpenAI Service in Azure AI Foundry