Databricks adds GPT‑5.5 and Codex to platform, routes model calls through governance layer

Databricks said Saturday that OpenAI’s GPT-5.5 and Codex are now available on its platform through Unity AI Gateway. (databricks.com) The company said customers can use GPT-5.5 for enterprise agents, document pipelines and business queries through Genie, while Codex handles coding workflows on the same platform. (databricks.com) Databricks said Unity AI Gateway governs both GPT-5.5 inference and Codex workflows through one control plane with permissions, cost controls, guardrails and observability. (databricks.com) The pitch is that companies can change models without rebuilding their compliance setup. Databricks said the gateway keeps access controls, audit logs and traffic policies consistent across providers. (databricks.com) That matters because newer AI systems do more than answer prompts. Databricks said agents now call models, databases, application programming interfaces and coding tools in one workflow, often touching sensitive company data at each step. (databricks.com) Databricks said every GPT-5.5 request or Codex interaction is logged with identity, tokens, latency and cost in Delta tables that customers own. It also said the gateway can enforce rate limits, detect personally identifiable information and block prompt injection. (databricks.com) The company has spent April widening that governance layer. On April 15, Databricks said AI Gateway would govern large language models, Model Context Protocol servers and external tools; on April 17, it added coding-agent support for products such as Cursor, Codex and Claude Code. (databricks.com 1) (databricks.com 2) Two days before this launch, Databricks said GPT-5.5 had set a new high score on its OfficeQA benchmark for document-heavy enterprise tasks. The company said GPT-5.5 scored 64.66% on OfficeQA Pro LLM with Oracle PDF and Web Search, up from GPT-5.4 at 57.14%. (databricks.com) In Databricks’ full-agent OfficeQA harness, which uses a Codex agent setup to find documents and compute answers, the company said GPT-5.5 scored 52.63% versus 36.10% for GPT-5.4, a 46% reduction in errors. (databricks.com) The immediate result is that Databricks customers can adopt OpenAI’s latest model and coding agent without moving governance into separate tools. Databricks is selling the model as replaceable and the control layer as the permanent part. (databricks.com 1) (databricks.com 2)