TaxBench Data Retrieval pass@1
Professional Work · 2026-05-04
Data Retrieval subset of TaxBench reporting pass@1 first-attempt performance on professional tax tasks.
Top models (higher is better)
| Model | Score |
|---|---|
| GPT-5.5 Pro | 74.5 |
| GPT-5.5 | 73.1 |
| Opus 4.6 | 71.9 |
| GPT-5.4 Pro | 67.0 |
| Gemini 3.1 Pro Preview | 66.0 |
| Opus 4.7 | 64.4 |
| GPT-5.2 Pro | 58.8 |
| Sonnet 4.6 | 58.8 |
| Sonnet 4.5 | 58.2 |
| GPT-5.4 | 56.9 |
| Grok 4.20 | 52.6 |
| Grok 4.1 Fast | 51.3 |
| GPT-5.2 | 48.7 |
| Gemini 3.1 Flash (benchmark label) | 42.8 |
| Gemini 2.5 Pro | 42.2 |