Language Models Show Significant Economic Value, Usage Lags Potential

Alexander Wan, Stephane Hatgis-Kessell, Tom\'as Aguirre, Percy Liang, Rishi Bommasani· July 23, 2026 View original

Summary

A new open-source evaluation suite, EconEvals, measures language models' economic value across US labor tasks, finding potential for substantial time savings in nearly half of all occupations. However, current usage of models like Claude significantly lags their potential, primarily due to privacy and proprietary system bottlenecks.

Researchers have developed EconEvals, an open-source evaluation framework designed to assess the economic utility of large language models (LLMs) across a wide spectrum of tasks, work activities, and occupations within the US economy. This suite improves upon existing benchmarks by significantly expanding coverage and reducing evaluation costs. The methodology incorporates real user queries and synthetic data to provide a comprehensive view of LLM capabilities. The study's findings indicate that current LLMs could save workers considerable time in at least half of their tasks across 47% of US occupations. Despite this substantial potential, observed usage of models like Claude remains low for 79% of tasks where significant time savings are predicted. This gap between potential and actual adoption is attributed primarily to concerns around data privacy and the limitations imposed by proprietary systems, highlighting key barriers to broader AI integration in the workforce.

Why it matters

Professionals can understand the quantified economic impact of LLMs on various job functions and identify areas where AI adoption is lagging despite high potential, informing strategic deployment and investment decisions.

How to implement this in your domain

  1. 1Evaluate internal workflows to identify tasks with high potential for LLM-driven time savings, especially in occupations identified by EconEvals.
  2. 2Develop pilot programs for LLM integration, focusing on tasks where current usage is low but potential savings are high.
  3. 3Address privacy and data security concerns by implementing secure LLM solutions or developing internal guidelines for sensitive data handling.
  4. 4Investigate open-source LLM alternatives to mitigate proprietary system bottlenecks and enhance customization.

Who benefits

ConsultingTechnologyHuman ResourcesGovernmentFinance

Key takeaways

  • EconEvals provides a new, cost-effective benchmark for assessing LLM economic value.
  • LLMs could save significant time in nearly half of US occupations.
  • Current LLM usage lags potential due to privacy and proprietary system issues.
  • Addressing these bottlenecks is crucial for maximizing AI's labor market impact.

Original post by Alexander Wan, Stephane Hatgis-Kessell, Tom\'as Aguirre, Percy Liang, Rishi Bommasani

"arXiv:2607.19375v1 Announce Type: cross Abstract: Language models perform economically valuable work, yet they are not currently assessed for how well they perform every economically valuable task. We introduce EconEvals as an open-source evaluation suite to measure capabilities…"

View on X

Originally posted by Alexander Wan, Stephane Hatgis-Kessell, Tom\'as Aguirre, Percy Liang, Rishi Bommasani on X · view source

Want to go deeper?

Turn these trends into skills with Learnijoy's hands-on AI & tech courses.

Explore courses

More in AI News & Tools

AI Engineering & DevToolsAI News & Tools

New Clustering Method Scales LLM Inference with Guardrails

A novel two-stage clustering algorithm enables efficient and scalable LLM inference by grouping inputs and using cluster representatives, guaranteeing minimal within-cluster similarity and exact categorical attribute matching at scale.

Longshaokan Wang, Wai Tsang Keung, Punit Ghodasara, Roman Wang, Ali Dashti, Francesc Moreno-NoguerJul 23, 2026
AI Engineering & DevToolsAI News & Tools

New Open-Access Dataset for Marine Engine Fault Diagnostics Released

Researchers have released the Marine Engine Fault Dataset, an open-access collection of multi-sensor time-series data from a three-cylinder marine diesel engine. This dataset includes both reference performance and controlled fault scenarios, providing a valuable benchmark for developing predictive maintenance and anomaly detection models in maritime machinery.

Ahmad BahooToroody, Oleksiy Bondarenko, Mohammad Mahdi Abaei, Niki Yoichi, Enrico ZioJul 23, 2026
AI ResearchAI Engineering & DevToolsAI News & Tools

Simulating Eutopia: Long-Term Fairness in AI Decision-Making

This paper introduces "Eutopia," a credit lending simulator, to study long-term fairness in AI-driven decision-makers (ADMs) by considering performative environments and downstream equity. The research formalizes wealth dynamics as a performative Markov Decision Process and demonstrates that learning with performative dynamics and fairness-aware utilities leads to better long-term efficiency, equity, and inclusivity.

Vedant Palit, Udvas Das, Brahim Driss, Debabrota BasuJul 23, 2026