GDPval

GDPval is a tool OpenAI built to work like a report card. It gives AI real tasks drawn from jobs like accounting, investing, or medicine—the kind of work people do at a desk all day—and then grades how usable the AI's output actually is in practice.

There have long been plenty of tests for measuring coding skill, but no solid way to measure how helpful AI actually is in non-coding work like accounting, law, or medicine. GDPval is an attempt to fill that gap. It's OpenAI's own benchmark for proving results as the company tries to expand tools that are already used by nearly everyone inside OpenAI out into other professions beyond the company.

One thing worth keeping in mind: GDPval itself isn't a product people use day to day. It's a measurement tool OpenAI uses internally to check performance and to use in promotional materials.

The article explains that "OpenAI uses its own benchmark called GDPval as the yardstick to close this gap." What's easy to confuse here is that GDPval isn't a product you actually use, like ChatGPT work or Codex—it's an evaluation tool OpenAI uses to measure and showcase the abilities of those kinds of products.