Skip to content
dotdock

GDPVAL: EVALUATING AI MODEL PERFORMANCE

2025

Publication cover

Open publication workspace · Sign in to read the full PDF.

AI-generated summary

GDPval is a new benchmark for evaluating AI model performance on real-world, economically valuable tasks, offering a more direct measure of AI's potential economic impact.

* **Comprehensive Task Coverage:** Evaluates AI on tasks across 44 occupations in the top 9 U.S. GDP sectors, mirroring real-world professional work.
* **Realistic Evaluation:** Tasks are derived from actual work products of experienced professionals, assessed through human expert comparison and an experimental automated grader.
* **Performance Insights:** Analyzes how frontier models perform over time, their strengths and weaknesses, and the impact of reasoning effort and scaffolding on their capabilities.

AI, performance evaluation, benchmark, real-world tasks, economic impact

Check the original publication for accuracy and context.