A benchmark map of the US economy
Mercor mapped 29 AI benchmarks to US job tasks and estimated that 90% of paid work hours have little or no coverage.
TL;DR
- Mercor mapped 29 AI benchmarks against 18,797 tasks in the US Labor Department’s O*NET taxonomy.
- Its map says 208 of 923 occupations have at least one task with five or more benchmark prompts, while about 90% of paid hours fall in minimally covered tasks.
- The estimates come from Mercor’s own methodology and have not been independently reproduced.
Mercor published a map comparing 29 AI benchmarks with 18,797 tasks in the US Department of Labor’s O*NET taxonomy. The company says it used a model to map benchmark prompts to occupational tasks. [1] [2]
The map reports that 208 of 923 occupations include a task with five or more prompts from an AI benchmark. It also estimates that roughly 90% of paid work hours are in tasks with no benchmark or only minimal coverage. [1] [2]
RuntimeWire described the map as exposing a gap between commonly reported model benchmarks and the broader set of work tasks. Mercor itself has a commercial role in supplying expert data and evaluation services, which is relevant context for a company-authored measurement. [1] [2]
The figures describe benchmark coverage, not the share of jobs that AI can or cannot perform. The methodology and estimates have not been independently reproduced, and the coverage map does not by itself measure workplace adoption or job displacement. [1] [2]
Why it matters
The map highlights a measurement gap in AI evaluation, but its company-produced estimates should be read as a benchmark inventory rather than a measure of automation or labor-market impact.
Editor's note
Mercor’s figures are self-reported estimates; RuntimeWire provides separate coverage, not independent replication.