Databricks releases GPT‑5.5, achieving 50% accuracy on OfficeQA Pro and 46% error reduction

Databricks announced GPT‑5.5, a new LLM that sets a new state‑of‑the‑art on the OfficeQA Pro benchmark, which tests parsing, retrieval, and grounded reasoning across complex enterprise documents. GPT‑5.5 achieved 50% accuracy, surpassing the 5.4 version, and reduced errors by 46% in agent‑harness workflows. The model shows a step‑function lift in parsing scanned PDFs and legacy files, and it retrieves relevant context more reliably, reducing unnecessary search detours. Databricks now makes GPT‑5.5 available through its AI Unity Gateway, allowing customers to use the model inside workflows built with AgentBricks and the Agent Supervisor API. The release is expected to improve reliability and efficiency for production agent systems that handle long‑context documents. The upgrade also brings improvements in orchestration across multi‑step tasks, better knowledge lift, and more reliable completion of complex workflows without additional supervision. Databricks anticipates widespread adoption of GPT‑5.5 in custom agent workflows, especially where parsing accuracy and error rates are critical.