Microsoft's MAI-Thinking-1 Reasoning Model Now Available for Azure AI Foundry Deployment
Microsoft has been working on its first in-house reasoning model, MAI-Thinking-1, which was shipped into Microsoft Foundry at Build 2026. This model is now available as a public preview inside Azure AI Foundry, but with surprisingly little plain-English documentation for Australian developers.
A tutorial has been created to close this gap and help deploy MAI-Thinking-1 in Azure AI Foundry. The process involves setting up an Azure subscription and Foundry project, checking region availability and deployment tiers, deploying the model via the Foundry portal or Azure CLI, retrieving the endpoint and API key, and authenticating against it.
MAI-Thinking-1 uses a Mixture-of-Experts (MoE) design, which means it activates only the parameters needed for a given request. This results in strong reasoning, math, and general intelligence at a fraction of the cost of other models.