OpenAI’s GPT-5.6 is now generally available on Azure Databricks, giving developers a direct path to run the latest frontier model inside the Databricks environment without stitching together separate infrastructure. You purchase access through Microsoft Foundry, then deploy via the Databricks Model Serving Endpoint. The integration means your data pipelines, feature engineering, and inference all live in one place.
The headline feature here is not just another model version landing. It’s that GPT-5.6 is the first OpenAI model to ship this way through Databricks, using Foundry as the purchasing layer and Databricks as the runtime. That matters because Databricks already handles Unity Catalog for governance, Lakebase for storage, and MLflow for experiment tracking. Dropping GPT-5.6 into that stack means security policies, lineage tracking, and cost controls apply automatically. No separate proxy, no custom middleware.
For practical impact, consider a team building a RAG pipeline over structured and unstructured data. They can call GPT-5.6 from the same notebook or pipeline that processes their source documents, feed the model endpoint with batching and rate limiting handled by Databricks, and log every invocation to Unity Catalog for audit. The Foundry integration also means teams can compare GPT-5.6 against other models available through Foundry in the same evaluation harness, using the managed VNET support that recently went GA for secure inference behind private endpoints.
To get started: purchase GPT-5.6 capacity through the Microsoft Foundry catalog, create a Model Serving Endpoint in Azure Databricks pointing to the Foundry-deployed model, and call it via the Databricks SDK or REST API from any workspace. Start with a small serving concurrency to manage costs – frontier models are expensive at scale. The Foundry-Databricks integration handles credential exchange automatically once both are configured in the same Azure subscription.
What this means: Microsoft is quietly building the rails for enterprise AI that don’t require teams to become infrastructure specialists. Foundry handles procurement and evaluation, Databricks handles runtime and governance, and the developer just writes prompts and pipelines. GPT-5.6 on Databricks is the first visible outcome of that deeper Foundry-Databricks integration, not the last.
Sources: Azure update announcement, Azure Databricks Model Serving docs, Foundry managed VNET for evaluations GA.