Azure AI just made GPT-4.1 fine-tuning faster and more accessible

Microsoft has expanded Azure AI Foundry’s fine-tuning tools with support for Direct Preference Optimization and more global regions. If you’ve been waiting to align models faster and smarter, this update delivers exactly that.

Fine-tuning GPT-4.1 models just got easier

First up is support for Direct Preference Optimization (DPO) in GPT-4.1 and 4.1-mini. This technique lets you fine-tune your model using side-by-side responses—just mark which output you like better, and DPO does the rest.

It’s faster and simpler than RLHF since it doesn’t need a reward model. And it still gives you alignment based on tone, safety, or your preferred style of communication.

Train models close to where you work

Microsoft is also expanding Global Training to 12 more regions. Now you can train your models closer to your data centers, cutting latency and increasing compliance. The new rollout includes:

• East US, East US 2, North Central US
• South Central US, West US, West US 3
• UK South, West Europe, Spain Central
• Sweden Central, Switzerland North and West

More regions mean more flexibility and scale for enterprise teams building across borders.

Now run fine-tuned models in the Responses API

Fine-tuning is one thing, but running those models smoothly matters just as much. That’s why Responses API now supports fine-tuned models too.

This means your agents can hold multi-turn conversations, call tools, keep context, and even show reasoning paths, without extra setup. You can also process queries in the background, or trigger web search and file lookups during a task.