NC Bench
A comprehensive benchmark for creative writing models.
Created by Novelcrafter.
This project is in early access and still work in progress.
Focused on Creativity
NC Bench is a cutting-edge benchmark for creativity in LLMs, with focus on creative writing, instruction following, utility, tooling, and language skills.
AI as an Assistant
We test how well models enhance the writing process through text manipulation, idea generation, summarization, and translation.
Comprehensive Testing
From generating quality prose to hallucination-free extraction, NC Bench puts AI models through their paces in all aspects of writing assistance.
Benchmark Overview
Tests
9Scenarios
62Models
93Results
55,512Samples per Run
9.27 runs/scen.Category Distribution
Shows the number of scenarios in each category. Some scenarios may be in multiple categories.
Creative writing (7)
Rule following (23)
Utility (30)
Mathematics (1)
Tooling (13)
Language (9)
Logic (8)
Top Results
Top Models
| 88.77% | MoonshotAI: Kimi K2.5 |
| 88.14% | Claude Opus 4.6 |
| 86.94% | o4 Mini High |
Creative writing
| 78.16% | o4 Mini High |
| 71.27% | Claude Opus 4.6 |
| 68.57% | Claude Opus 4.5 |
Rule following
| 92.07% | o4 Mini High |
| 87.60% | o4 Mini |
| 87.44% | MoonshotAI: Kimi K2.5 |
Utility
| 95.00% | Claude Sonnet 4 |
| 92.83% | Claude Opus 4.6 |
| 91.83% | Claude 3.5 Sonnet (new) |
Mathematics
| 100.00% | Claude 3.5 Sonnet |
| 100.00% | Claude 3.7 Sonnet |
| 100.00% | Gemini Flash 1.5 |
Tooling
| 100.00% | Gemini 3 Pro (Preview) |
| 100.00% | MoonshotAI: Kimi K2.5 |
| 99.23% | Gemini 2.5 Pro |
Language
| 98.13% | Hermes 3 405B |
| 96.28% | DeepSeek-V2 Chat |
| 95.68% | Z.AI GLM 4.5 |
Logic
| 94.38% | Phi-3 Mini 128k |
| 93.75% | GPT-4o Mini (temp=0) |
| 93.75% | Claude Sonnet 4 |