ConStory-Bench: leaderboard
Long story generation benchmark measuring cross-scene consistency bugs using Consistency Error Density over 2,000 generated stories.
Metric: CED errors per 10K words (lower is better). Source: github.com. Source: picrew.github.io. Saturation forecast: Around January 2027. 66 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 | 0.11 | #98 |
| 2 | Gemini 2.5 Pro | 0.3 | #152 |
| 3 | Gemini 2.5 Flash | 0.31 | #245 |
| 4 | Claude Sonnet 4.5 | 0.52 | #142 |
| 5 | GLM-4.6 | 0.53 | #252 |
| 6 | Qwen 3 32B | 0.54 | #443 |
| 7 | Ring-1T | 0.54 | #319 |
| 8 | DeepSeek V3.2 Exp | 0.54 | #234 |
| 9 | Qwen 3 235B A22B (Thinking) | 0.56 | #313 (Qwen 3 235B A22B) |
| 10 | GLM-4.5 | 0.6 | #275 |
| 11 | Grok 4 | 0.67 | #181 |
| 12 | Ling-1T | 0.7 | #348 |
| 13 | GPT-4o (2024-11-20) | 0.71 | #381 |
| 14 | Step-3 | 0.85 | #365 |
| 15 | Qwen 3 Next 80B A3B (Thinking) | 0.96 | #320 (Qwen 3 Next 80B A3B) |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=constory-bench · How It Works · Data refreshed daily, snapshot 2026-10-20.