Benchmarks / Tax Agent Bench
Tax Agent Bench
Built by Vals AI · released 4 Sep 2026
Vals AI's test of whether an agent can do the research work of a corporate tax associate. Each task is one written US tax question, mostly federal corporate income tax, with the rest in partnership tax, state and local tax, transfer pricing, and employment tax. The agent researches with five tools (web search, authority lookup by citation, document fetch, retrieval, calculator) and submits one written answer with citations.
Frontier
49.2%
Accuracy
Claude Fable 5.1 · harness: Vals Tax Agent Bench harness
8 Oct 2026 · Source: Vals AI (benchmark maintainers)
First on Accuracy, the all-pass score scaled by citation quality (page data 49.194, stderr 3.606), at about $13.18 per task and about 29 minutes per question, compute effort max. Also first on Weighted Pass (77.642), the partial-credit score that the atlas showed as the headline until 2026-10-10. Claude Opus 5.5 (45.57%) and Gemini 4 Argon (44.56%) follow. Date is the page's "updated" field.
The best result is at 20–60% of the ceiling, or at 50–100% of human parity.
What it measures
Accuracy on a private test set of 193 expert-written questions, graded by an LLM judge (GPT 5.4) against an expert rubric of 3 to 89 weighted checks per question (mean 23.6). A question scores only when every check passes; the score is then scaled by the share of cited sources that resolve to real authoritative documents (a fully correct answer scores 70% to 100%). Each task has a three-hour limit. Vals also reports Weighted Pass, a partial-credit score of rubric points earned that is zeroed when a must-pass check fails (about 40% of checks are must-pass).
Accuracy: Share of the 193 private test questions answered fully correctly (every rubric check passed), with each correct answer scaled by citation quality (0.7 + 0.3 x the share of cited sources that resolve). Higher is better.
Status compares the frontier with a ceiling of 100%.
Facts
- Grain
- Task level: bounded tasks with a clear spec
- Environment
- api-tools, documents
- Grading
- llm-judge, rubric
- Tasks
- 391
- Human reference
- Tax professionals wrote and reviewed every question, gold answer, and rubric. Reference answers average 601 words. No human score is published.
- Contamination
- The test set is private. Vals reports scores only on the test set.
- Reuse
- 391 questions in total: 5 public samples in the MIT-licensed repository, 193 validation questions available for license, and a 193-question test set that Vals says will stay private. (cite-only)
Limits to keep in mind
- Grading is by an LLM judge (GPT 5.4) against expert rubrics. Until a change between 2026-09-24 and 2026-10-08 the headline was the partial-credit score now called Weighted Pass, where Claude Fable 5.1 scores 77.64%; under the strict all-pass Accuracy no model reaches 50%. Source
- Standard errors are about 3.5 to 3.6 points, so the top four models (44.56% to 49.19%) overlap. Source
- Tools are fixed by the Vals harness, and the private set can only be run by Vals. Some document fetches fail on blocked or missing pages (3.8% of tool calls failed in the original 18-model analysis set). Source
- The questions are research questions with a written answer. They do not cover preparing a return end to end or client work, so the score speaks to tax research, not all tax work. Source
Recorded results
A selection that shows the frontier over time. The full leaderboard is at the source.
Timeline
Where it sits in the atlas
Work ladder: Direct evidence for Finance and accounting. On the work ladder it counts as 49.2% of tasks completed. How the ladder works
Work it measures (O*NET work activities): Advise others on legal or regulatory matters; Calculate financial data.
Go to the source
- Website www.vals.ai
- Full leaderboard www.vals.ai
- Code github.com
Last checked 10 Oct 2026 against 3 primary sources. See an error? Tell us.
How to cite
Credit the original work first: Tax Agent Bench by Vals AI.
Then, if you used this page:
Can Agents Work. "Tax Agent Bench: frontier results and sources." https://canagentswork.com/benchmarks/vals-tax-agent-bench/ (accessed 2026-10-11). CC BY 4.0.@misc{caw-vals-tax-agent-bench,
title = {{Tax Agent Bench: frontier results and sources}},
author = {{Can Agents Work}},
year = {2026},
howpublished = {\url{https://canagentswork.com/benchmarks/vals-tax-agent-bench/}},
note = {Accessed 2026-10-11. CC BY 4.0}
}