Databricks brings GPT-5.5 to enterprise agent workflows

Databricks is making GPT-5.5 available for customer agent workflows. This integration follows the model's achievement of a new state of the art on OfficeQA Pro, Databricks’ benchmark for complex enterprise document tasks. This advancement is set to enhance how businesses handle intricate document processes.

OfficeQA Pro evaluates how AI models manage parsing, retrieval, and grounded reasoning across various document types, including scanned PDFs, legacy files, and long-context documents. These tasks often pose significant challenges for production agent systems due to the complexity and potential for error.

GPT-5.5 demonstrated a substantial improvement, reducing errors by 46% compared to GPT-5.4 in the agent-harness setting. It also became the first model to surpass 50% accuracy on the OfficeQA Pro benchmark, highlighting its enhanced capability in handling enterprise-level document challenges.

The most significant gains from GPT-5.5 were observed in parsing-heavy workflows. Earlier models struggled with correctly extracting information from older documents and scanned PDFs. However, GPT-5.5 shows a remarkable improvement in these areas, ensuring greater accuracy from the initial stages of document processing.

Improvements were also noted in the orchestration of multi-step tasks. Previous models sometimes engaged in inefficient search detours, which GPT-5.5 has largely overcome. This leads to more reliable retrieval of relevant context and more efficient completion of complex workflows without extensive supervision.

Databricks customers can now access GPT-5.5 through AI Unity Gateway, using it within workflows built with AgentBricks and the Agent Supervisor API. In these systems, GPT-5.5 handles the orchestration of parsing, retrieval, and execution across specialized agents, offering a robust solution for enterprise automation.