Databricks has partnered with OpenAI on GPT-5.5, OpenAI's latest frontier model optimized for agentic enterprise work, complex document reasoning, and long-horizon coding tasks. GPT-5.5 now powers Codex, OpenAI's coding agent. Databricks evaluated the model on its OfficeQA benchmark — built from 89,000 pages of U.S. Treasury Bulletins — where GPT-5.5 scored 64.66% (up ~13% from GPT-5.4) in the document-retrieval-assisted setting, and 52.63% (a 46% error reduction) in the full end-to-end agent workflow setting. GPT-5.5 is coming soon to the Databricks platform.
Sort: