Databricks Integrates GPT-5.5 for Enterprise Agent Workflows

Databricks has integrated GPT-5.5 into its enterprise agent workflows to improve the reliability of complex document processing. This integration matters because GPT-5.5 is the first model to surpass 50% accuracy on the OfficeQA Pro benchmark, significantly reducing errors in parsing and reasoning across legacy and scanned documents.

State-of-the-Art Performance on OfficeQA Pro

GPT-5.5 has established a new state-of-the-art (SOTA) on OfficeQA Pro, a benchmark specifically designed to evaluate how models handle parsing, retrieval, and grounded reasoning across scanned PDFs and legacy files. In agent-harness settings, GPT-5.5 reduced errors by 46% compared to GPT-5.4, becoming the first model to exceed 50% accuracy on this benchmark.

According to Research Engineer Arnav Singhvi, the model exhibits a "step-function lift" in parsing older documents and scanned PDFs, where previous models like GPT-5.4 often failed to extract digits and numbers correctly. This improvement in parsing is critical because extraction errors in these workflows typically cascade downstream, altering the entire trajectory of the agent's work.

Improved Orchestration and Reliability

Beyond parsing, GPT-5.5 demonstrates improved orchestration across multi-step tasks. While GPT-5.4 occasionally performed "unnecessary search detours" that led to inefficient trajectories, GPT-5.5 is more reliable at retrieving relevant context and completing complex workflows without requiring additional supervision.

Enterprise Production Integration

Databricks is making GPT-5.5 available through the AI Unity Gateway. Customers can utilize the model within workflows built with AgentBricks and the Agent Supervisor API, where GPT-5.5 acts as the orchestrator for parsing, retrieval, and execution across specialized agents.

As Research Engineer Arnav Singhvi notes, "GPT-5.5 has been great in terms of knowledge lift. It’s a step size function change in terms of doing knowledge work for us."