Databricks brings GPT-5.5 to enterprise agent workflows

Databricks has made GPT-5.5 available for enterprise AI agent workflows, where the model achieved a new benchmark record by reaching 50% accuracy on OfficeQA Pro (a test measuring how well AI systems handle complex business document tasks like parsing scanned PDFs and legacy files). Compared to the previous GPT-5.4 model, GPT-5.5 reduced errors by 46% and showed major improvements in parsing old documents and managing multi-step tasks without unnecessary detours.