Atlas

Benchmarks

← All benchmarks

TaxCalcBench TY25 (web search)

Professional Work · 2026-06-22

TaxCalcBench TY25's 50 federal and state tax-return cases with web search enabled. Its headline strict score is the percentage of cases for which every evaluated line exactly matches the expected return; runs without external tools are scored under TaxCalcBench TY25.

Top models (higher is better)

ModelScore
GPT-6 Sol66.0
Claude Opus 5.564.0
GPT-5.6 Sol64.0
GPT-6.1 Sol64.0
GPT-6 Astra62.0
GPT-5.558.0
Claude Fable 5.148.0
Claude Opus 544.0
Claude Sonnet 5.544.0
Muse Wonder42.0
Claude Fable 538.0
Opus 4.832.0
Kimi K332.0
GPT-6 Luna30.0
Muse Spark 1.326.0
Gemini 3.7 Flash24.0
Claude Haiku 5.520.0
Gemini 3.6 Flash18.0
Muse Spark 1.216.0
Mistral Large 4 Preview4.0
Loading Atlas data…