Tax Agent Bench — Forms & Filings
Professional Work · 2026-09-23
Weighted rubric and citation credit on 30 tax forms-and-filings questions, using the parent evaluation protocol.
Top models (higher is better)
| Model | Score |
|---|---|
| DeepSeek V4.1 Flash 1017 | 79.4 |
| Muse Wonder Full | 76.6 |
| MiniMax M3.1 Flash | 75.3 |
| Claude Fable 5.1 | 73.2 |
| GPT-5.6 Sol | 72.5 |
| Grok 4.6 | 72.1 |
| Muse Spark 1.3 | 71.7 |
| GLM-5.3 | 71.7 |
| Mistral Large 4 Preview | 71.7 |
| Qwen3.8 Max (0902) | 71.5 |
| Muse Wonder | 69.8 |
| GPT-5.6 Terra | 68.7 |
| Kimi K3 | 68.6 |
| Hy4 Preview | 68.5 |
| Gemini 4 Argon | 67.3 |
| Claude Opus 5 | 67.2 |
| Ember-1 | 65.9 |
| Gemini 4 Carbon | 65.8 |
| Sonnet 4.6 | 65.6 |
| Sonnet 5 | 65.2 |
| Gemini 3.8 Flash | 65.1 |
| Grok 4.5 | 64.2 |
| Claude Fable 5 | 64.0 |
| Grok 4.7 | 63.8 |
| GPT-6 Luna | 62.9 |
| Qwen3.8 27B | 62.4 |
| Kimi K3.1 Pro | 61.9 |
| Claude Haiku 5.5 | 61.8 |
| GLM-5.3 Flash | 61.4 |
| GPT-6 Astra | 60.8 |
| Opus 4.7 | 60.4 |
| GPT-5.5 | 59.9 |
| Muse Spark 1.2 | 59.5 |
| DeepSeek-V4-Flash-0731 | 59.4 |
| MiMo-V2.6-Pro | 59.1 |
| Step 5 Preview | 58.8 |
| GPT-5.6 Luna | 58.5 |
| GPT-6.1 Sol | 57.7 |
| DeepSeek V4.1 Flash | 56.8 |
| Opus 4.8 | 55.8 |