Among Microsoft’s portfolio of developer tools is Azure AI Studio. The platform debuted as a preview earlier this year for building and deploying generative AI applications. While the company is using artificial intelligence to make coding easier through its Copilot offering, it also wants to help programmers build their own software, providing a resource hub containing models and tools one might need to turn their ideas into reality.

This week, at the Microsoft Build conference, the company is making Azure AI Studio generally available.

This so-called “pro-code” platform is different from when it was first announced. Several new features are present, but perhaps the most significant is that OpenAI’s flagship model, GPT-4o, is now available in Azure AI Studio and as an API.

GPT-4o in Azure AI Studio

If you’re going to build an AI-powered app, it’s important to have access to the latest top-of-the-line models, right? That would include OpenAI’s GPT-4o, which joins more than 1,600 other models from Mistral, Meta, Hugging Face, Databricks, Cohere, Nvidia, Snowflake, and, of course, Microsoft. Developers can use this multimodal foundational model to incorporate text, image and audio processing into their apps to provide generative and conversational AI experiences.

To support developers who want to build their own custom copilot applications, Microsoft offers alternative workflows to make the process easier and faster. This includes support for Azure Developer CLI, or AZD, and the new AI Toolkit for VS Code.

The former is an open-source tool providing pre-built components and templates developers can use instead of worrying about infrastructure concerns and automating their apps’ testing and deployment.

The latter provides developers with a choice of how they want to fine-tune their models—locally or in the cloud—before they are deployed to Azure or other platforms using container images.

Both of these integrations are in early preview with no date for general availability announced yet.

New MaaS model additions

Microsoft is also expanding its Models-as-a-Service (MaaS) offering. The program, which was announced in 2023, allows developers to rent inference APIs and host fine-tuning through a pay-as-you-go plan. It’s akin to the cloud computing financial model. The idea is to eliminate the need for dedicated virtual machines.

Today, TimeGen-1 from Nixtila and Core42 JAIS are being added to this collection, though only as a preview. They join Mistral-7B and Meta’s Llama 2, but they won’t be alone for long. Microsoft says it will soon add models from AI21, Bria AI, Gretel Labs, NTT Data, Stability AI and Cohere. Developers can access these MaaS models through Arize and ClearML.

What would a development platform be without a packaged set of tools? Azure AI Studio now has an AI toolchain, including seamless data integration, prompt orchestration and system evaluation.

There are also controls for prompt flow allowing for the management of workflow orchestration both for multimodal models and MaaS. This includes using images in conversations and models such as Llama 3, Mistral Large and Cohere Command R+.

Finally, Azure AI Studio is equipped with AI tracing, debugging and monitoring, making it possible for developers to better understand AI workflows—what’s working and what’s not—and analyze an app’s performance to track key token usage along with quality and operational metrics once the software has been deployed. All of these features are in preview today.