Databricks brings GPT-5.5 to enterprise agent workflows

Databricks has made GPT-5.5 available for customer agent workflows following the model's performance on OfficeQA Pro, a benchmark designed for complex enterprise document tasks. OfficeQA Pro evaluates how models handle parsing, retrieval, and grounded reasoning across scanned PDFs, legacy files, and long-context documents. In the agent-harness setting, GPT-5.5 reduced errors by 46 percent compared to GPT-5.4 and became the first model to surpass 50 percent accuracy on the benchmark. According to the company, Databricks saw the largest gains in parsing-heavy workflows where earlier models struggled to extract digits correctly from older documents and scanned PDFs. Additionally, GPT-5.5 demonstrated improvements in multi-step task orchestration, avoiding the unnecessary search detours seen in prior models and proving more reliable at retrieving relevant context without additional supervision. Databricks now offers GPT-5.5 through the AI Unity Gateway, allowing customers to use the model inside workflows built with AgentBricks and the Agent Supervisor API. In these systems, the model orchestrates parsing, retrieval, and execution across specialized agents to support custom enterprise agent workflows.