Harvey HLAB - Criteria Pass Rate - Tax
Professional Work · 2026-05-06
The Tax practice-area split of Harvey's Legal Agent Benchmark held-out evaluation criteria pass rate. This reports pooled rubric criteria passed within that split and is distinct from the all-pass task-score split.
Top models (higher is better)
| Model | Score |
|---|---|
| Gemini 4 Argon | 93.3 |
| Muse Spark 1.2 | 93.0 |
| Muse Spark 1.3 | 93.0 |
| Muse Spark 1.1 | 91.3 |
| Kimi K3 | 87.6 |
| Muse Wonder Full | 87.3 |
| Step 5 Preview | 87.3 |
| Ember-1 | 86.6 |
| Gemini 3.7 Flash | 86.6 |
| MiMo-V2.6-Flash | 86.3 |
| Sonnet 4.6 | 86.0 |
| GLM-5.3 | 86.0 |
| Mistral Large 4 Preview | 86.0 |
| GLM-5.2 | 85.6 |
| DeepSeek V4.1 Flash 1017 | 85.0 |
| Grok 4.6 | 85.0 |
| MiniMax M3 | 85.0 |
| Claude Opus 5 | 84.6 |
| Opus 4.7 | 84.3 |
| Grok 4.5 | 84.3 |
| Claude Sonnet 5.5 | 83.9 |
| Gemini 3.8 Flash | 83.9 |
| Hy4 Preview | 83.9 |
| MiMo-V2.6-Pro | 83.9 |
| Qwen3.8-Max | 83.6 |
| DeepSeek-V4-Flash-0731 | 82.9 |
| Muse Wonder | 82.6 |
| Sonnet 5 | 82.3 |
| DeepSeek V4.1 Flash | 82.3 |
| Grok 4.7 | 81.9 |
| DeepSeek V4 Pro 0813 | 81.3 |
| Qwen3.8 27B | 80.9 |
| Opus 4.8 | 80.6 |
| GPT-6.1 Sol | 78.9 |
| DeepSeek-V4-Pro | 77.9 |
| MiMo-V2.5 | 77.9 |
| Gemini 3.6 Flash | 77.6 |
| Qwen3.6 Plus (2026-04-02) | 77.6 |
| Qwen3.7-Max (source unspecified) | 77.6 |
| Inkling-Small | 76.6 |