原文 · 未翻译
Databricks brings GPT-5.5 to enterprise agent workflows | OpenAI
Databricks brings GPT-5.5 to enterprise agent workflows | OpenAI
May 15, 2026
Databricks brings GPT‑5.5 to enterprise agent workflows
GPT‑5.5 set a new state of the art on OfficeQA Pro, Databricks’ benchmark for complex enterprise agent tasks.
Contact sales
Company size: Enterprise
Region: North America
Industry: Technology
Products: Codex
50%
Accuracy on the OfficeQA Pro benchmark (state of the art)
46%
Reduction in errors on the OfficeQA Pro benchmark vs. GPT-5.4
Listen to article
Audio 1
Databricks is making GPT‑5.5 available for customer agent workflows after the model established a new state of the art on OfficeQA Pro, the company’s benchmark for complex enterprise document tasks.
OfficeQA Pro evaluates how models handle parsing, retrieval, and grounded reasoning across workflows involving scanned PDFs, legacy files, and long-context documents, tasks that frequently break production agent systems.
In the agent-harness setting, GPT‑5.5 reduced errors by 46% compared to GPT‑5.4 and became the first model to surpass 50% accuracy on OfficeQA Pro.
“Codex with 5.5 is now state-of-the-art amongst all the agents and models out there.”
–Arnav Singhvi, Research Engineer
Video 1
00:00
SOTA performance on OfficeQA Pro
OfficeQA Pro includes large numbers of scanned or legacy enterprise documents, where small extraction errors during parsing can cascade downstream through the rest of the workflow. “Once you can’t extract a certain digit or number, that changes the entire trajectory of what the agent works with,” Singhvi explains.
Databricks saw the largest gains from GPT‑5.5 in these parsing-heavy workflows. “Earlier models like 5.4 were unable to parse all the digits correctly, but it seems like 5.5 has a step-function lift in parsing older documents and scanned PDFs,” Singhvi says.
The team also saw improvements in orchestration across multi-step tasks. “One thing that we saw with 5.4 is sometimes it would go on these unnecessary search detours, and that would cause very inefficient trajectories,” Singhvi says.
Compared to earlier models, GPT‑5.5 was more reliable at retrieving relevant context and completing complex workflows without additional supervision.