GDPval
OpenAI
220 real work tasks across 44 occupations, with reference files, to test AI on economically valuable output.
Tasks written by professionals, each with a text prompt and supporting files such as spreadsheets and slide decks. One row is one task. Model deliverables are graded by experts rather than by automatic checks. A canary string is included to keep it out of training data.
Access
- Cost
- FreeNo charge.
- License
- License not statedNot stated in the Hub record
- Sign-up
- No sign-up
- How you get it
- API and Bulk downloadParquet, PDF, XLSX, PPTX
What's in it
- One row is
- One row per occupation task
- Covers
- Not geographic
- Years
- 2025
- Updated
- One-time release
- Size
- 220 tasks across 44 occupations
Quality
Novel in using real work products and expert grading; small and the card includes sensitive content and AI-generated people.
- Small sample per occupation
- Grading requires experts
- Includes NSFW and political tasks by design
Connects to
- SpreadsheetBench: Task-level benchmark focused on one office skill
- Anthropic Economic Index: Maps AI use to occupational tasks, a complement to task-based testing
Questions it could answer
- Which occupations have the most tasks needing attached files?
- How do task prompts differ across sectors?