Story Theory Bench — leaderboard
Creative narrative generation benchmark: LLMs write short stories incorporating 10 required elements (characters, objects, concepts, attributes, motivations). Scored 50% programmatic + 50% LLM judge across 34 tasks.
Metric: Score (%). Source: github.com. Status: saturated. 35 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-5.4 | 99.6 |
| 2 | GLM-5 | 99.6 |
| 3 | Mercury 2 | 99.1 |
| 4 | Qwen 3.5 27B | 98.9 |
| 5 | Gemini 3.1 Flash Lite (Preview) | 98.4 |
| 6 | Claude Opus 4.6 | 98.1 |
| 7 | DeepSeek V3.2 | 92.2 |
| 8 | Claude Opus 4.5 | 90.9 |
| 9 | Claude Sonnet 4.5 | 90.1 |
| 10 | Step 3.5 Flash | 90.1 |
| 11 | Claude Sonnet 4 | 89.6 |
| 12 | O3 | 89.5 |
| 13 | GLM-4.7 | 88.8 |
| 14 | Kimi K2 (Thinking) | 88.7 |
| 15 | Gemini 3 Flash (Preview) | 88.3 |
Interactive version: theaggregate.ai/benchmark?slug=story-theory-bench · How the rankings work · Data refreshed daily, snapshot 2026-07-22.