Surge AI releases DAYJOB: Tax benchmark for professional tax agents

• Surge AI launched DAYJOB: Tax on October 27, 2026, a benchmark designed to assess AI agents on end-to-end professional tax workflows including compliance, advisory, and dispute resolution.
• The benchmark consists of 50 environments across 8 tax practice areas, requiring agents to process 1,789 workspace files (median 30 per task) and achieve a 95% rubric score to pass.
• Claude Sonnet 5.5 achieved the highest pass rate of 10.8%, followed by Claude Fable 5.1 at 9.6% and Claude Opus 5.5 at 9.2%, with Sonnet and Opus 5.5 positioned on the high-performance Pareto frontier.
• Analysis of agent failures reveals that frontier models struggle with professional judgment under incomplete information, such as missing statutory deadlines or overlooking legal marital status in retirement account divisions.

Benchmark Composition

• Tasks are designed by tax experts from PwC, Deloitte, Exxon, Baker Tilly, Moss Adams, and EY, undergoing a 3-stage expert review for accuracy and solvability.
• Input data includes unstructured evidence such as PDFs, Excel sheets, IRS notices, ledgers, and site visit logs, extending beyond simple calculation to verify if records support tax treatment.

Differentiation and Findings

• Unlike TaxBench or TaxCalcBench, DAYJOB evaluates the discovery of missed issues and end-to-end responsibility in complex scenarios like partnership loss analysis and estate filings.
• The results indicate that while current models can execute individual steps, they have not yet reached consistent performance in real-world tax workflows requiring sustained professional judgment.