DramaChain Bench - Storyboard Design: leaderboard
DramaChain Bench (Tencent Hunyuan, 2026) scores an automated short-drama production chain stage by stage; this is its text stage, in which a language model turns each episode script into a shooting storyboard, 533 items over 60 episodes of 20 dramas. An agentic judge that averages GPT-5.6 Sol and a Claude Fable 5 variant scores eight leaf dimensions on a 1-5 rubric (event coverage, dialogue fidelity, executability, restraint in additions, narrative flow, shooting rhythm, emotional expression and audiovisual style), and the board is their mean, higher is better, with 3.0 as the usable line. The judge reproduces the three-annotator human model ranking at PLCC 0.936 on the annotated models; models added after annotation were scored by the judge alone, including models from the judges' own families.
Metric: Composite rubric score (1-5; mean of 8 leaf dimensions). Source: arxiv.org. Saturation forecast: Around February 2027. 16 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.6 Sol (Max) | 4.4 | #29 (GPT-5.6 Sol) |
| 2 | Claude Fable 5 (Max) | 4.24 | #26 (Claude Fable 5) |
| 3 | GPT-5.5 (xHigh) | 4.23 | #48 (GPT-5.5) |
| 4 | Kimi K3 | 4.22 | #53 |
| 5 | Claude Opus 5 (Max) | 4.22 | #20 (Claude Opus 5) |
| 6 | Claude Opus 4.8 (Max) | 4.17 | #56 (Claude Opus 4.8) |
| 7 | Qwen 3.8 Max | 4.08 | #55 |
| 8 | Grok 4.5 | 3.92 | #78 |
| 9 | Hy3 | 3.89 | #193 |
| 10 | Seed 2.1 Pro | 3.89 | #133 |
| 11 | GLM-5.2 | 3.79 | #137 |
| 12 | Gemini 3.6 Flash (High) | 3.78 | #72 (Gemini 3.6 Flash) |
| 13 | Gemini 3.1 Pro (Preview) | 3.71 | #91 |
| 14 | Kimi K2.6 | 3.64 | #173 |
| 15 | Qwen 3.7 Max | 3.48 | #142 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=dramachain-bench-storyboard-design · How It Works · Data refreshed daily, snapshot 2026-11-12.