Databricks has integrated GPT-5.5 into its enterprise agent workflows after the model achieved state-of-the-art performance on OfficeQA Pro, the company's benchmark for complex document processing tasks. The model secured 50% accuracy on the challenging evaluation suite, marking a significant advancement in enterprise AI capabilities.

OfficeQA Pro tests how models handle parsing, retrieval, and grounded reasoning across workflows involving scanned PDFs, legacy files, and long-context documents. These tasks frequently break production agent systems due to small extraction errors that cascade downstream. Arnav Singhvi, Research Engineer at Databricks, noted that parsing inaccuracies fundamentally change agent trajectories.

GPT-5.5 delivered a 46% reduction in errors compared to GPT-5.4, with the most substantial gains in parsing-heavy workflows involving older documents. The model also showed improved orchestration across multi-step tasks, eliminating unnecessary search detours that plagued earlier versions. Databricks now offers GPT-5.5 through AI Unity Gateway for use with AgentBricks and Agent Supervisor API.

Customers can deploy the model to supervise custom agent workflows, leveraging its enhanced knowledge work capabilities. The step-function improvement in parsing scanned documents and orchestrating complex tasks represents a practical advancement for enterprise automation where document processing accuracy directly impacts business outcomes.