TaxBench: leaderboard
TaxBench evaluates AI models on real-world tax tasks from Rivet's active tax workflows, spanning tax knowledge and judgment, tax calculations, and agentic data-retrieval question answering.
Metric: Mean pass^5 (computed) (imported). Source: www.rivet.tax. Source: benchmarklist.com. Saturation forecast: Around July 2027. 16 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.5 Pro | 29.27 | #17 |
| 2 | GPT-5.4 Pro (xHigh) | 27.03 | #32 (GPT-5.4 Pro) |
| 3 | GPT-5.5 | 24.43 | #31 |
| 4 | Claude Opus 4.6 | 21.37 | #63 |
| 5 | Gemini 3.1 Pro (Preview) | 20.1 | #58 |
| 6 | Grok 4.1 Fast (Reasoning) | 19.63 | #217 (Grok 4.1 Fast) |
| 7 | GPT-5.2 Pro | 17.77 | #61 |
| 8 | Grok 4.20 (Reasoning) | 15.2 | #175 (Grok 4.20) |
| 9 | Claude Opus 4.7 | 14.37 | #49 |
| 10 | Grok 4.20 | 12.23 | #175 |
| 11 | Claude Sonnet 4.6 | 11.2 | #90 |
| 12 | GPT-5.4 | 9.33 | #79 |
| 13 | Gemini 2.5 Pro | 9 | #152 |
| 14 | Claude Sonnet 4.5 | 8.03 | #142 |
| 15 | GPT-5.2 | 4.6 | #110 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=taxbench · How It Works · Data refreshed daily, snapshot 2026-10-20.