Skip to content
Source: ZDNET

🧠Microsoft Lets Customers Fine-Tune Its Frontier AI Models With Their Own Data

Your private data can now shape Microsoft's AI models

TL;DR

Microsoft will let customers customize its frontier AI models with their own data, run securely in Azure via Licensed Weights. The pitch is privacy, capability, and cost. Enterprises that want control over their models should start talking to their account reps now.

Microsoft is letting enterprises fine-tune its frontier AI models on their own data. The program, called Licensed Weights, runs the customized models securely inside Azure. Companies that want in have to go through their Microsoft account reps to get started. If you're building internal tools on Azure OpenAI or similar, this changes the buy-versus-build math. Teams with sensitive data in regulated industries now have a path to a tuned model without shipping data to a third party. Microsoft says it will only use customer data to train its own models if customers explicitly agree. The launch came at Ignite alongside several new MAI models. MAI Code 2 reportedly delivers 98% of GPT 6.1 Sol's performance at 40% of the cost, and about 5% of the cost of Opus 5.5. MAI Image 2.7 leads the category on Artificial Analysis, ahead of Nano Banana Pro and several GPT image models on price and quality. Worth watching: Anthropic and OpenAI still hold the overwhelming majority of enterprise AI spend. Whether Licensed Weights pulls buyers away from them depends on how cheap and fast the tuned models end up in practice.

Microsoft Lets Customers Fine-Tune Its Frontier AI Models With Their Own Data — ZDNET

Key Points

1

Microsoft's Licensed Weights initiative lets customers customize frontier MAI models with their own data, running them securely inside Azure rather than on external infrastructure.

Keep reading — it's free

Enter your email to keep reading — plus the best of AI & tech, daily. Free, forever.

Also get
or

Already a member? Sign in

microsoftazureopen-weight-aienterprise-aimai-models

Comments

Subscribe to join the conversation...

Be the first to comment