Run OpenAI Models Directly on Databricks

OpenAI and Databricks have partnered to make OpenAI latest models, including GPT-5, natively available on the Databricks Data Intelligence Platform. For the first time, enterprises can run GPT-5 on their data within Databricks in a secure and governed way. Users can call GPT-5 from within SQL, as an API endpoint within Model Serving, or via Agent Bricks. This enables organizations to confidently build and deploy agents optimized on their data with the quality and data protection that their most critical workloads demand.

GPT-5 + Databricks = Smarter AI Agents

With Databricks, you can securely tap into OpenAI GPT-5 model and build agents that reason on your enterprise data. Agent Bricks then helps you iterate on domain-specific tasks so your agents continuously improve in accuracy and quality.

OpenAI GPT-5 delivers major advances in instruction following and agentic tool use, allowing it to reliably execute multi-step requests, coordinate across different tools, and adapt to shifting context. Compared to prior generations, GPT-5 more faithfully follows directions and gets more of the work done end-to-end, whether the task is straightforward or requires complex, evolving reasoning.

Databricks builds on this foundation by ensuring governed access to models like GPT-5, so enterprises can use them securely and compliantly with their data. On top of this, Agent Bricks provides a built-in evaluation loop: it measures the accuracy of models like GPT-5 on domain-specific tasks, then tunes and optimizes them for higher quality. This continuous improvement cycle ensures enterprises get production-ready agents that deliver measurable quality in their specific business context.

Examples include:

• Reasoning agents for healthcare decision support, financial risk analysis, contract review, and logistics planning, where GPT-5’s extended reasoning mode helps analyze long documents and multi-step calculations.
• Productivity agents for summarization, business writing, knowledge management, and customer support, benefiting from GPT-5’s stronger writing quality and lower factual error rates.
• Coding agents for debugging large systems, modernizing legacy applications, and generating production-ready code

“From consumer to enterprise solutions, our mission is to deliver AI-powered insights with speed, security and trust,” said Connie Cheung, VP of Engineering at Experian. “The partnership between Databricks and OpenAI is an exciting step forward, enabling us to build, evaluate and deploy trusted AI applications on our data with Databricks while tapping into the latest OpenAI models."

“At Mastercard, we are focused on delivering AI solutions that make commerce safer, smarter, and more personalized. Agentic solutions also make our internal operations stronger by automating processes and optimizing performance across our systems,” said Greg Ulrich, Chief AI and Data Officer at Mastercard. “For any use case, AI agents come down to three things: quality, scale, and trust. This partnership between Databricks and OpenAI enables us to build on the strong foundations we’ve established with each firm and provides us the opportunity to build trusted AI agents that harness the latest OpenAI models, delivered with the speed, security and scale of the Databricks platform.”

Secure, Governed Access to GPT-5

Enterprises can use OpenAI models like GPT-5 from day one via API or SQL with security and governance built in. Mosaic AI Gateway delivers this foundation by enforcing access controls, monitoring, and policies on every request. The result is that organizations can scale GPT-5 confidently across the business without compromising compliance or reliability.

Easily connect GPT-5 to reason over your data on Databricks with one SQL command:

This launch also marks the beginning of a deep engineering collaboration between OpenAI and Databricks. Our teams will co-develop feedback loops, model improvements, and Databricks-specific optimizations to ensure OpenAI models perform best on enterprise tasks.

Getting Started with GPT-5 on Databricks

OpenAI models will soon be available in Foundation Model APIs on Databricks. The easiest way to begin is with AI Playground, where you can try out GPT-5 in just a few clicks.

From there, you can:

• Use SQL: Call GPT-5 with ai_query to classify tickets, summarize documents, or extract insights directly from your tables.
• Use APIs: Call GPT-5 programmatically to integrate into apps and workflows.

Next Steps

The following OpenAI models will be natively available soon in Databricks across AWS, Azure, and GCP:

• GPT-5
• GPT-5 mini
• GPT-5 nano

Get access to the Public Preview by reaching out to your account team. And join our Virtual Event, “The future of AI: Build Agents that Work”, with Sam Altman and Ali Ghodsi.