OpenAI has partnered with the Pacific Northwest National Laboratory to develop DraftNEPABench, a new benchmark evaluating how AI coding agents can accelerate federal permitting processes. The collaboration between OpenAI and PNNL's PermitAI team aims to modernize infrastructure reviews that currently take years to complete.

Working with 19 subject matter experts, the teams designed the benchmark to assess AI performance on National Environmental Policy Act workflows, including drafting environmental impact statements. Testing across 18 federal agencies revealed that generalized coding agents could reduce NEPA document drafting time by up to 15%, potentially saving 1 to 5 hours per subsection.

The benchmark demonstrates AI's potential to handle document-heavy government processes by reading technical reports, cross-checking information across multiple sources, and drafting structured analyses that meet regulatory requirements. Beyond efficiency gains, this work advances better interfaces for experts, moving from static PDFs to dynamic web-based reports and interactive visualizations that make validation easier for human reviewers.