TaxCalcBench
Models compute complete US federal tax returns line by line from taxpayer data, with no tax engine. A return counts only when every evaluated line is exact.
Tax Year 2024 edition. The leaderboard still offers a TY 2024 view next to v2.
In the authors’ words
Can AI file your taxes? Not yet. Calculating US personal income taxes is a task that requires building an understanding of vast amounts of English text and using that knowledge to carefully compute results. We propose TaxCalcBench, a benchmark for determining models' abilities to calculate personal income tax returns given all of the necessary information. Our experiment shows that state-of-the-art models succeed in calculating less than a third of federal income tax returns even on this simplified sample set. Our analysis concludes that models consistently misuse tax tables, make errors in tax calculation, and incorrectly determine eligibility. Our findings point to the need for additional infrastructure to apply LLMs to the personal income tax calculation task.
- Level
- Expert Specialists with years of training, often over hours of work
- Scored
- Checked automatically A program checks the result against an answer key, unit tests or the expected final state.
- Answers
- Short answer A number, word or short string matched to a key.
2 models share the top score of 62.8% among models released by Mar 2026
Best score per model from the benchmark's own leaderboard, read on 2 Nov 2026. Where a model ran with several settings or harnesses, we keep its best.
The newest model with a score here came out in Mar 2026, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:
- GPT-6 Astra
- Claude Opus 5.5
- GPT-6.1 Terra
- GPT-6.1 Sol
- and 15 more
- 62.8 GPT-5.4
- 62.8 GPT-5.4 Pro
- 52.9 Claude Opus 4.6
- 49.0
Gemini 3.1 Pro - 41.7 GPT-5
- 41.2 GPT-5.2 Pro
- 37.3 Claude Sonnet 4.6
- 36.3 Gemini 3 Pro
- 36.3 Claude Opus 4.5
- 33.8 GPT-5.2
- 32.4 Gemini 2.5 Pro
- 31.4 Claude Sonnet 4.5
- 28.4 Claude Opus 4.1
- 27.4 Claude Opus 4
- 26.0 Gemini 2.5 Flash
- 23.0 Claude Sonnet 4
- 13.7 Claude Haiku 4.5
- 1 GPT-5.4 62.8%
- 2 GPT-5.4 Pro 62.8%
- 3 Claude Opus 4.6 52.9%
- 4 Gemini 3.1 Pro 49%
- 5 GPT-5 web search 41.7%
- 6 GPT-5.2 Pro 41.2%
- 7 Claude Sonnet 4.6 37.3%
- 8 Gemini 3 Pro 36.3%
- 9 Claude Opus 4.5 36.3%
- 10 GPT-5.2 33.8%
- 11 Gemini 2.5 Pro May 2025 32.4%
- 12 Claude Sonnet 4.5 31.4%
- 13 Claude Opus 4.1 28.4%
- 14 Claude Opus 4 27.4%
- 15 Gemini 2.5 Flash May 2025 26%
- 16 Claude Sonnet 4 23%
- 17 Claude Haiku 4.5 web search 13.7%
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | GPT-5.4 | 62.8% | Mar 2026 | Official leaderboard | |
| 2 | GPT-5.4 Pro | 62.8% | Mar 2026 | Official leaderboard | |
| 3 | Claude Opus 4.6 | 52.9% | Feb 2026 | Official leaderboard | |
| 4 | Gemini 3.1 Pro | 49% | Feb 2026 | Official leaderboard | |
| 5 | GPT-5 | 41.7% | web search | Aug 2025 | Official leaderboard |
| 6 | GPT-5.2 Pro | 41.2% | Dec 2025 | Official leaderboard | |
| 7 | Claude Sonnet 4.6 | 37.3% | Feb 2026 | Official leaderboard | |
| 8 | Gemini 3 Pro | 36.3% | Nov 2025 | Official leaderboard | |
| 9 | Claude Opus 4.5 | 36.3% | Nov 2025 | Official leaderboard | |
| 10 | GPT-5.2 | 33.8% | Dec 2025 | Official leaderboard | |
| 11 | Gemini 2.5 Pro May 2025 | 32.4% | May 2025 | Official leaderboard | |
| 12 | Claude Sonnet 4.5 | 31.4% | Sep 2025 | Official leaderboard | |
| 13 | Claude Opus 4.1 | 28.4% | Aug 2025 | Official leaderboard | |
| 14 | Claude Opus 4 | 27.4% | May 2025 | Official leaderboard | |
| 15 | Gemini 2.5 Flash May 2025 | 26% | May 2025 | Official leaderboard | |
| 16 | Claude Sonnet 4 | 23% | May 2025 | Official leaderboard | |
| 17 | Claude Haiku 4.5 | 13.7% | web search | Oct 2025 | Official leaderboard |
4 models have held number one since its release in Jul 2025
Lab claims count from the day the card came out, independent runs from the model’s release.
Scores over time
All the scores we have, one dot each, so a model tested at several settings appears more than once.
No lab model card reports it yet
We read the headline results of 479 launch posts from 51 labs, and none of them list this benchmark.
Versions
We switch the top row to each new version when it comes out. Open it for one row per version. An old version stays in play while people still test it, so whoever first takes it past 85% gets credit for closing it out.
More in Law, finance and medicine
- HealthBench Professional In use 525 physician-authored tasks from real clinician chats (care consult, documentation, research) with physician-adjudicated rubrics. Claude Sonnet 5.5 69.2%
- MedXpertQA-MM In use Models answer expert-level medical exam questions that combine images with clinical records and test results. Scoring is accuracy.
Qwen 3.8 Max 80.4% - HealthBench In use 5,000 realistic health conversations graded against physician-written rubrics. A length-adjusted variant penalizes verbosity. GPT-5 67.2%
- KBL In use Models answer Korean legal knowledge and reasoning tasks and Korean bar exam questions. It has 7 knowledge tasks, 4 reasoning tasks and 2,510 bar exam items.
Solar Pro 4 77.5% - KorMedMCQA In use Models answer 7,469 multiple-choice questions from Korean licensing exams for doctors, nurses, pharmacists and dentists, held from 2012 to 2024. Solar Pro 4 93.2%
- PRBench In use Model answers 1,100 expert-written law and finance questions. LLM judge grades against weighted expert rubrics of 10-30 criteria each.
Muse Spark 1.3 61.6%