Mercor has published a US Benchmark Map that measures how well existing AI benchmarks cover the work Americans actually do. The company's headline finding: roughly 90% of US working hours are spent on tasks with little or no coverage from current benchmarks, which makes it very hard to judge how good AI models are at real jobs.
The map is built on the US Department of Labor's O*NET taxonomy, which splits the workforce into 923 occupations and 18,797 tasks. Mercor says it is a first step toward a continuously updated picture of AI capabilities across the economy, and it warns the results are provisional.
What the US Benchmark Map found
Mercor started from a list of 29 benchmarks chosen for quality and economic relevance, then used an LLM judge to tag each benchmark prompt with the O*NET tasks it matches. A task counts as covered only if at least five prompts map to it. The company validated the tagging with human expert labels on its own APEX benchmarks, a subset of the 29.
On that basis, only 208 of 923 occupations have a task with at least five benchmark prompts, and 6 of 22 job families have no such occupations at all.
- Only 208 of 923 O*NET occupations have a task covered by at least 5 benchmark prompts
- 6 of 22 job families have no covered occupations
- About 90% of US working hours go to tasks with little or no benchmark coverage
- Coverage is 6x more likely for O*NET tasks rated AI-exposed than unexposed
