Step 2 of 4 · ProjectsBenchmarks
PolicyBench
Benchmark of how well frontier AI models estimate household tax and benefit outcomes without tools.
What it does
A benchmark of how well frontier AI models estimate household tax and benefit outputs without tools: it samples households from PolicyEngine's US dataset under 2026 rules, asks each model for numbers and explanations, and scores them. It is a Python package with a dashboard, needing API keys for the model providers and PolicyEngine US for reference answers. US only; released under the Unlicense.
Where it fits on the map
Similar projects
Of the same kind, for a shared jurisdiction or tax area. Not a ranking.
- HSCodeCompExpert-level benchmark for HS code classification agents on e-commerce products, from Alibaba's MarcoPolo.
- Legora transfer pricing caseEvaluation case for AI: draft a transfer pricing Master File from case documents, graded on 175 criteria.
- SARAStatutory reasoning benchmark built on the US Internal Revenue Code.
- TaxCalcBenchBenchmark that tests frontier AI models on US tax return calculations.
Maintain this project?
Link to this page from your README, so readers find the facts, the notes and where it fits. The badge is a small SVG served from this site.
[](https://tax-oss.com/projects/policybench/)