Categories
NC Bench evaluates models across 8 categories and 23 subcategories.
Category Distribution
Shows the number of scenarios in each category. Some scenarios may be in multiple categories.
Tooling (15)
Creative Writing (18)
Language (9)
Utility (32)
Reasoning (21)
Text Editing (18)
Rule Following (12)
Hallucination (30)
Creative Writing
18 scenarios · 6 subcategories
Top Models
| 91.17% | GPT-5.4 (Reasoning) |
| 91.04% | GPT-5.6 Sol |
| 91.03% | GPT-5.6 Sol (Reasoning) |
Subcategories
| 82.84% | AI-isms |
| 68.36% | Prose Variety |
| 80.23% | Dialogue |
| 89.13% | Purple Prose |
| 86.84% | Mechanical Style |
| 78.71% | Clichés |
Tooling
15 scenarios · 1 subcategory
Top Models
| 100.00% | GPT-5.6 Sol (Reasoning) |
| 100.00% | Claude Opus 4.6 (Reasoning) |
| 100.00% | Qwen3.7 Max |
Subcategories
| 96.41% | XML |
Language
9 scenarios · 2 subcategories
Top Models
| 100.00% | Qwen3.6 Max Preview |
| 100.00% | Claude Sonnet 4.6 |
| 100.00% | DeepSeek-V2 Chat |
Subcategories
| 85.10% | Comprehension |
| 89.14% | Generation |
Utility
32 scenarios · 5 subcategories
Top Models
| 99.91% | Gemini 3.1 Pro (Preview) |
| 99.54% | Qwen3.7 Max |
| 99.26% | Claude Opus 4.8 (Reasoning) |
Subcategories
| 68.16% | Word Counting |
| 91.44% | Sentence Counting |
| 97.17% | Paragraph Counting |
| 76.42% | Structural Counting |
| 98.39% | Data Extraction |
Reasoning
21 scenarios · 2 subcategories
Top Models
| 96.86% | GPT-5.6 Sol (Reasoning) |
| 94.89% | GPT-5.4 (Reasoning) |
| 94.45% | Grok 4.5 (Reasoning, High) |
Text Editing
18 scenarios · 3 subcategories
Top Models
| 99.13% | Claude Sonnet 4 |
| 99.02% | Claude Sonnet 4.5 |
| 98.90% | Z.AI GLM 5.2 (Reasoning, High) |
Subcategories
| 85.77% | Transformation |
| 94.72% | Preservation |
| 98.73% | Structural Integrity |
Rule Following
12 scenarios · 1 subcategory
Top Models
| 95.76% | Qwen3.7 Max |
| 92.04% | Gemini 3.5 Flash (Reasoning) |
| 91.21% | Gemini 3.1 Pro (Preview) |
Subcategories
| 63.71% | Constraint Adherence |
Hallucination
30 scenarios · 3 subcategories
Top Models
| 99.53% | o4 Mini High |
| 99.32% | Z.AI GLM 5 Turbo |
| 99.06% | Claude Opus 4.6 (Reasoning) |
Subcategories
| 75.58% | False Positives |
| 97.74% | Content Invention |
| 99.65% | Output Corruption |