FailedMoney / bills

Have a model compute a full federal tax return

Tools: Frontier LLMs (Gemini, GPT, Claude)

Setup, access, approval gate, failure notes
Setup / prompt
Give the model a realistic taxpayer's documents and ask for a complete Form 1040.
Access required
Tax documents only (benchmark)
HITL checkpoint
None in the benchmark, which is the point: shows what unsupervised output looks like
Outcome & failure notes
Strict-correct returns (every evaluated line matching) were around a third in the first run and only just over half for the best models in later editions. Models confuse tax years and miss forms. Note the benchmark is published by a tax-software company.
Evidence type
Open benchmark

Source: Column Tax (TaxCalcBench) · 2025-07 · Checked 2026-10-19