Leaderboard
Tax Agent Bench
Agentic research on US tax questions, about 60% federal corporate income tax. The rest covers partnership, state and local, transfer pricing and employment tax. Agents use tools and must answer with citations.
Best reported
49.7
Qwen 3.8 Max 0902
accuracy % (all-pass x citation factor)
29 Oct 2026
Tax Agent Bench: score vs cost
Score against cost per test. Up and to the left is better.
Updated 29 Oct 2026
Tax Agent Bench 2026-10-29
accuracy % (all-pass x citation factor), higher is better. 10 models.
Updated 29 Oct 2026
- Qwen 3.8 Max 090249.7
- Claude Fable 5.149.2
- Gemini 4 Carbon48.9
- Claude Opus 5.545.6
- Claude Opus 545.5
- Gemini 4 Argon44.6
- DeepSeek V4.1 Flash 101744.4
- Wonder Full43
- Muse Spark 1.3 Max42.4
- Claude Sonnet 5.542.0
Tax Agent Bench: cost
cost per test item (USD), lower is better. 10 models.
Updated 29 Oct 2026
- Muse Spark 1.3 Max0.390
- DeepSeek V4.1 Flash 10170.460
- Wonder Full2.1
- Gemini 4 Argon3.5
- Qwen 3.8 Max 09024.1
- Claude Opus 55.1
- Gemini 4 Carbon6.3
- Claude Sonnet 5.59.3
- Claude Fable 5.113.2
- Claude Opus 5.515.1
Tax Agent Bench 2026-10-29
accuracy % (all-pass x citation factor), higher is better. 10 models.
Updated 29 Oct 2026
- 1Qwen 3.8 Max 090249.7
- 2Claude Fable 5.149.2
- 3Gemini 4 Carbon48.9
- 4Claude Opus 5.545.6
- 5Claude Opus 545.5
- 6Gemini 4 Argon44.6
- 7DeepSeek V4.1 Flash 101744.4
- 8Wonder Full43
- 9Muse Spark 1.3 Max42.4
- 10Claude Sonnet 5.542.0
| 1 | Qwen 3.8 Max 0902 | Qwen | 49.7 | test set | Vals.ai leaderboard Official board, 29 Oct 2026 |
|---|---|---|---|---|---|
| 2 | Claude Fable 5.1 | Anthropic | 49.2 | test set | Vals.ai leaderboard Official board, 29 Oct 2026 |
| 3 | Gemini 4 Carbon | 48.9 | test set | Vals.ai leaderboard Official board, 29 Oct 2026 | |
| 4 | Claude Opus 5.5 | Anthropic | 45.6 | test set | Vals.ai leaderboard Official board, 29 Oct 2026 |
| 5 | Claude Opus 5 | Anthropic | 45.5 | test set | Vals.ai leaderboard Official board, 29 Oct 2026 |
| 6 | Gemini 4 Argon | 44.6 | test set | Vals.ai leaderboard Official board, 29 Oct 2026 | |
| 7 | DeepSeek V4.1 Flash 1017 | DeepSeek | 44.4 | test set | Vals.ai leaderboard Official board, 29 Oct 2026 |
| 8 | Wonder Full | Not stated | 43 | test set | Vals.ai leaderboard Official board, 29 Oct 2026 |
| 9 | Muse Spark 1.3 Max | Not stated | 42.4 | test set | Vals.ai leaderboard Official board, 29 Oct 2026 |
| 10 | Claude Sonnet 5.5 | Anthropic | 42.0 | test set | Vals.ai leaderboard Official board, 29 Oct 2026 |
