TaxEval v2 - Stepwise Reasoning
Professional Work · 2025-04-18
The Stepwise Reasoning split of TaxEval v2. This child benchmark separates a source-reported subtask or subtrack from the parent aggregate so scores at different grains do not share one benchmark_slug.
Top models (higher is better)
| Model | Score |
|---|---|
| Muse Spark 1.1 | 92.6 |
| Muse Spark | 89.8 |
| Sonnet 4.6 | 88.4 |
| Claude Opus 5 | 88.3 |
| Claude Fable 5 | 88.1 |
| Grok 3 | 87.9 |
| Sonnet 5 | 87.2 |
| Opus 4.6 | 87.2 |
| Inkling | 87.1 |
| Qwen3.7-Max | 87.1 |
| Kimi K2.5 | 86.9 |
| Gemini 3.5 Flash | 86.8 |
| Opus 4.7 | 86.8 |
| Opus 4.8 | 86.8 |
| Kimi K2.6 | 86.6 |
| GPT-5.6 Terra | 86.5 |
| GPT-4.1 | 86.3 |
| Gemini 3.6 Flash | 86.2 |
| Kimi K3 | 86.1 |
| Opus 4.5 | 85.9 |
| GPT-5.6 Luna | 85.8 |
| GPT-5 Mini | 85.7 |
| Qwen3 Max Preview | 85.5 |
| Gemini 3 Flash Preview | 85.4 |
| MiniMax M3 | 85.4 |
| Qwen3.6 Plus (2026-04-02) | 85.3 |
| Qwen3 Max (rolling alias) | 85.3 |
| Sonnet 4.5 | 85.2 |
| GPT-5.2 | 85.2 |
| o3 | 85.1 |
| O4 Mini | 85.1 |
| Gemini 3.1 Pro Preview | 85.0 |
| GPT-5.1 | 84.9 |
| Grok 4.20 | 84.9 |
| Claude 3.7 Sonnet | 84.8 |
| GLM-5.2 | 84.6 |
| Grok 4.1 Fast | 84.6 |
| GPT-5.6 Sol | 84.5 |
| Mistral Large 3 675B Instruct 2512 | 84.5 |
| Nemotron 3 Ultra 550B A55B | 84.5 |