openaienmodel: gpt-5-mini-2025-08-07
Databricks brings GPT-5.5 to enterprise agent workflows
Key Points
- 50% accuracy on OfficeQA Pro
- 46% error reduction vs GPT-5.4
- Available via AI Unity Gateway for AgentBricks
Summary
Databricks is making GPT-5.5 available for enterprise agent workflows after the model set a new state of the art on OfficeQA Pro. GPT-5.5 achieved 50% accuracy on OfficeQA Pro and reduced errors by 46% versus GPT-5.4, with the largest gains in parsing scanned and legacy documents, long-context retrieval, and multi-step orchestration. The model is accessible through AI Unity Gateway and integrates with AgentBricks and the Agent Supervisor API for production agent pipelines.
Key Points
- Performance: 50% accuracy on OfficeQA Pro; 46% reduction in errors vs GPT-5.4.
- Improvements concentrated on parsing extraction (scanned PDFs, legacy files), retrieval relevance, and orchestration across multi-step workflows.
- Production access: available via AI Unity Gateway; recommended usage inside AgentBricks workflows and supervised by the Agent Supervisor API.
Recommended actions for engineers
- Evaluate GPT-5.5 on your parsing-heavy benchmarks (scanned PDFs, legacy formats) and compare extraction error rates to GPT-5.4.
- Run controlled A/B experiments and staged rollouts; monitor parsing/extraction accuracy, retrieval relevance, and end-to-end workflow failure modes.
- Use the Agent Supervisor API patterns (supervision, fallbacks, and validation hooks) to catch downstream cascades from small parsing errors.
- Validate long-context behavior and orchestration efficiency on representative multi-step tasks before full production rollout.