TaxCalcBench v2
Muse Wonder 77.5% leads since Oct 2026 Not in model cards Models read W-2, 1099 and other PDFs and compute full federal and state returns line by line. A return counts only when every evaluated line is exact.
v2 adds realistic PDF inputs, state returns and more complex cases, so scores differ from v1. 50 returns. Fireworks lists 14 models.
Made by Column Tax Released Jun 2026 Finance All tasks public
Last model added Oct 2026
Scores refreshed 2 Nov 2026
- 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.
26
Models scored on its own leaderboard
Muse Wonder leads with 77.5%
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. Every run is in the next chart.
- 77.5 Muse Wonder
- 72.0 GPT-6 Sol
- 70.0 Claude Opus 5.5
- 70.0 GPT-6.1 Sol
- 68.0 GPT-5.6 Sol
- 68.0 GPT-6 Astra
- 64.0 GPT-5.5
- 61.0
Gemini 3.8 Flash - 54.0 Claude Fable 5.1
- 50.0 Claude Opus 5
- 48.0 Claude Sonnet 5.5
- 44.0 Claude Fable 5
- 38.0 Kimi K3
- 36.0 Claude Opus 4.8
- 36.0 GPT-6 Luna
- 32.0 Muse Spark 1.3
- 26.0 Gemini 3.7 Flash
- 22.0 Claude Haiku 5.5
- 20.0 Gemini 3.6 Flash
- 20.0 Muse Spark 1.2
- 6.00 Claude Sonnet 5
- 4.00
DeepSeek V4.1 Flash - 4.00
Mistral Large 4 - 2.00 Gemini 3.5 Flash
- 0.0 Gemini 3.1 Pro
- 0.0 GLM-5.3
- 1 Muse Wonder web search 77.5%
- 2 GPT-6 Sol web search 72%
- 3 Claude Opus 5.5 web search 70%
- 4 GPT-6.1 Sol web search 70%
- 5 GPT-5.6 Sol web search 68%
- 6 GPT-6 Astra web search 68%
- 7 GPT-5.5 web search 64%
- 8 Gemini 3.8 Flash web search 61%
- 9 Claude Fable 5.1 web search 54%
- 10 Claude Opus 5 web search 50%
- 11 Claude Sonnet 5.5 web search 48%
- 12 Claude Fable 5 web search 44%
- 13 Kimi K3 web search 38%
- 14 Claude Opus 4.8 web search 36%
- 15 GPT-6 Luna web search 36%
- 16 Muse Spark 1.3 web search 32%
- 17 Gemini 3.7 Flash web search 26%
- 18 Claude Haiku 5.5 web search 22%
- 19 Gemini 3.6 Flash web search 20%
- 20 Muse Spark 1.2 web search 20%
- 21 Claude Sonnet 5 6.00%
- 22 DeepSeek V4.1 Flash 4.00%
- 23 Mistral Large 4 web search 4.00%
- 24 Gemini 3.5 Flash 2.00%
- 25 Gemini 3.1 Pro 0%
- 26 GLM-5.3 0%
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | Muse Wonder | 77.5% | web search | Oct 2026 | Official leaderboard |
| 2 | GPT-6 Sol | 72% | web search | Sep 2026 | Official leaderboard |
| 3 | Claude Opus 5.5 | 70% | web search | Sep 2026 | Official leaderboard |
| 4 | GPT-6.1 Sol | 70% | web search | Sep 2026 | Official leaderboard |
| 5 | GPT-5.6 Sol | 68% | web search | Jul 2026 | Official leaderboard |
| 6 | GPT-6 Astra | 68% | web search | Sep 2026 | Official leaderboard |
| 7 | GPT-5.5 | 64% | web search | Apr 2026 | Official leaderboard |
| 8 | Gemini 3.8 Flash | 61% | web search | Sep 2026 | Official leaderboard |
| 9 | Claude Fable 5.1 | 54% | web search | Sep 2026 | Official leaderboard |
| 10 | Claude Opus 5 | 50% | web search | Jul 2026 | Official leaderboard |
| 11 | Claude Sonnet 5.5 | 48% | web search | Sep 2026 | Official leaderboard |
| 12 | Claude Fable 5 | 44% | web search | Jun 2026 | Official leaderboard |
| 13 | Kimi K3 | 38% | web search | Jul 2026 | Official leaderboard |
| 14 | Claude Opus 4.8 | 36% | web search | May 2026 | Official leaderboard |
| 15 | GPT-6 Luna | 36% | web search | Sep 2026 | Official leaderboard |
| 16 | Muse Spark 1.3 | 32% | web search | Sep 2026 | Official leaderboard |
| 17 | Gemini 3.7 Flash | 26% | web search | Aug 2026 | Official leaderboard |
| 18 | Claude Haiku 5.5 | 22% | web search | Oct 2026 | Official leaderboard |
| 19 | Gemini 3.6 Flash | 20% | web search | Jul 2026 | Official leaderboard |
| 20 | Muse Spark 1.2 | 20% | web search | Aug 2026 | Official leaderboard |
| 21 | Claude Sonnet 5 | 6.00% | Jun 2026 | Official leaderboard | |
| 22 | DeepSeek V4.1 Flash | 4.00% | Sep 2026 | Official leaderboard | |
| 23 | Mistral Large 4 | 4.00% | web search | Oct 2026 | Official leaderboard |
| 24 | Gemini 3.5 Flash | 2.00% | May 2026 | Official leaderboard | |
| 25 | Gemini 3.1 Pro | 0% | Feb 2026 | Official leaderboard | |
| 26 | GLM-5.3 | 0% | Aug 2026 | Official leaderboard |
The best score, 77.5%, costs $2.83 a task
Runs on the cost frontier score higher than any run that's cheaper. The costs are the leaderboard's own estimates, so compare them only within this benchmark.
OpenAI Anthropic Meta
Cost frontier
One model at each effort level
| Model | Effort | Harness | Score | Cost per task | Output tokens |
|---|---|---|---|---|---|
| Muse Wonder | web search | 77.5% | $2.83 | ||
| GPT-6 Sol | web search | 72% | $0.34 | ||
| Claude Opus 5.5 | web search | 70% | $3.50 | ||
| GPT-6.1 Sol | web search | 70% | $0.33 | ||
| GPT-5.6 Sol | web search | 68% | |||
| GPT-6 Astra | web search | 68% | $0.78 | ||
| GPT-5.5 | web search | 64% | |||
| Gemini 3.8 Flash | web search | 61% | $0.99 | ||
| Claude Fable 5.1 | web search | 54% | $5.27 | ||
| Claude Opus 5 | web search | 50% | $2.09 | ||
| Claude Sonnet 5.5 | web search | 48% | $2.38 | ||
| Claude Fable 5 | web search | 44% | |||
| GPT-6 Astra | 44% | $1.03 | |||
| Claude Fable 5.1 | 42% | $2.34 | |||
| Claude Opus 5.5 | 42% | $1.13 | |||
| Kimi K3 | web search | 38% | $0.59 | ||
| GPT-6.1 Sol | 38% | $0.19 | |||
| Claude Opus 4.8 | web search | 36% | |||
| GPT-6 Luna | web search | 36% | $0.24 | ||
| Muse Spark 1.3 | web search | 32% | $0.53 | ||
| Claude Fable 5 | 30% | ||||
| GPT-5.6 Sol | 30% | ||||
| GPT-6 Sol | 28% | $0.24 | |||
| Gemini 3.7 Flash | web search | 26% | $0.30 | ||
| GPT-5.5 | 24% | ||||
| Claude Haiku 5.5 | web search | 22% | $0.34 | ||
| Muse Wonder | 22% | $0.74 | |||
| Claude Opus 4.8 | 20% | ||||
| Gemini 3.6 Flash | web search | 20% | $0.53 | ||
| Claude Opus 5 | 20% | $0.58 | |||
| Muse Spark 1.2 | web search | 20% | $1.11 | ||
| Claude Sonnet 5.5 | 18% | $0.31 | |||
| Gemini 3.7 Flash | 14% | $0.07 | |||
| GPT-6 Luna | 14% | $0.03 | |||
| Gemini 3.6 Flash | 8.00% | ||||
| Claude Sonnet 5 | 6.00% | ||||
| Kimi K3 | 6.00% | ||||
| Muse Spark 1.2 | 6.00% | $0.06 | |||
| Gemini 3.8 Flash | 6.00% | $0.16 | |||
| Muse Spark 1.3 | 6.00% | $0.07 | |||
| DeepSeek V4.1 Flash | 4.00% | $0.06 | |||
| Mistral Large 4 | web search | 4.00% | $0.11 | ||
| Gemini 3.5 Flash | 2.00% | ||||
| Claude Haiku 5.5 | 2.00% | $0.02 | |||
| Gemini 3.1 Pro | 0% | ||||
| GLM-5.3 | 0% | $0.42 | |||
| Mistral Large 4 | 0% | $0.05 |
4 models have held number one since its release in Jun 2026
Lab claims count from the day the card came out, independent runs from the model’s release.
OpenAI Meta
Scores over time
All the scores we have, one dot each, so a model tested at several settings appears more than once.
On the benchmark’s own leaderboard
Best so far
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.
Google DeepMind OpenAI Anthropic Meta
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%