Tax Agent Bench — Numbers & Calculations
Professional Work · 2026-09-23
Weighted rubric and citation credit on 30 quantitative tax questions, using the parent evaluation protocol.
Top models (higher is better)
| Model | Score |
|---|---|
| Qwen3.8 Max (0902) | 88.2 |
| Claude Fable 5.1 | 88.0 |
| Gemini 4 Argon | 87.6 |
| Claude Opus 5 | 85.2 |
| GLM-5.3 | 84.5 |
| MiniMax M3.1 Flash | 84.5 |
| Kimi K3 | 84.0 |
| Claude Fable 5 | 83.4 |
| Gemini 3.8 Flash | 83.1 |
| Gemini 4 Carbon | 83.0 |
| Muse Spark 1.3 | 81.8 |
| GPT-5.6 Terra | 81.6 |
| GPT-5.6 Sol | 81.4 |
| Hy4 Preview | 81.0 |
| GLM-5.3 Flash | 81.0 |
| Sonnet 4.6 | 80.5 |
| Grok 4.6 | 79.8 |
| DeepSeek V4.1 Flash 1017 | 79.5 |
| Gemini 3.7 Flash | 78.8 |
| Opus 4.8 | 78.6 |
| Mistral Large 4 Preview | 78.4 |
| DeepSeek-V4-Flash-0731 | 78.0 |
| Gemini 3.6 Flash | 77.9 |
| GPT-6 Astra | 77.9 |
| Ember-1 | 77.7 |
| GPT-6.1 Sol | 77.3 |
| Opus 4.7 | 77.0 |
| Claude Haiku 5.5 | 76.8 |
| GPT-5.5 | 76.8 |
| Grok 4.7 | 76.7 |
| DeepSeek V4.1 Flash | 76.6 |
| Muse Spark 1.2 | 76.3 |
| Muse Wonder Full | 76.3 |
| MiMo-V2.6-Pro | 75.5 |
| Muse Wonder | 74.4 |
| GPT-6 Luna | 73.8 |
| Kimi K3.1 Pro | 72.9 |
| Gemini 3.5 Flash | 72.5 |
| Grok 4.5 | 72.5 |
| Step 5 Preview | 72.1 |