Tax Agent Bench — Citation-Adjusted Fact-Pattern All-Pass
Professional Work · 2026-10-08
Citation-adjusted all-pass credit on 43 held-out fact-pattern questions, under the October 8 headline scoring protocol.
Top models (higher is better)
| Model | Score |
|---|---|
| DeepSeek V4.1 Flash 1017 | 55.8 |
| Claude Fable 5.1 | 51.2 |
| Claude Opus 5 | 48.8 |
| Muse Wonder Full | 46.5 |
| Muse Spark 1.3 | 45.8 |
| Ember-1 | 39.5 |
| Gemini 4 Argon | 39.5 |
| Muse Wonder | 39.5 |
| Claude Fable 5 | 39.4 |
| MiMo-V2.6-Flash | 39.0 |
| GPT-5.6 Terra | 37.2 |
| Gemini 3.8 Flash | 34.9 |
| GLM-5.3 | 34.9 |
| Grok 4.7 | 34.9 |
| Qwen3.8 27B | 34.9 |
| Kimi K3 | 32.6 |
| Muse Spark 1.2 | 32.6 |
| MiniMax M3.1 Flash | 31.9 |
| Step 5 Preview | 31.9 |
| Mistral Large 4 Preview | 31.8 |
| GPT-6.1 Sol | 30.2 |
| MiMo-V2.6-Pro | 30.2 |
| Hy4 Preview | 30.2 |
| Claude Haiku 5.5 | 30.2 |
| GPT-6 Luna | 29.5 |
| DeepSeek V4 Pro 0813 | 28.1 |
| GPT-5.5 | 27.9 |
| GPT-6 Astra | 27.9 |
| Grok 4.6 | 27.8 |
| GLM-5.3 Flash | 27.8 |
| DeepSeek V4.1 Flash | 27.2 |
| Sonnet 5 | 27.1 |
| Muse Spark Open | 26.7 |
| Opus 4.8 | 25.6 |
| GPT-5.6 Luna | 25.6 |
| Inkling | 24.6 |
| DeepSeek-V4-Flash-0731 | 24.2 |
| Sonnet 4.6 | 23.3 |
| GPT-5.6 Sol | 23.3 |
| Grok 4.5 | 20.9 |