STAGE - Cross-Scene Reasoning (Hybrid RAG): leaderboard
STAGE (Tian et al., 2026) Task II: 5,010 questions about how characters' situations, knowledge and relationships change across scenes of 151 English and Chinese film screenplays, answered from scenes fetched by combined dense and lexical retrieval. Pass@5 is the percentage of questions where at least one of five generations (temperature 0.2) is judged correct by DeepSeek-V4-Pro against the reference answer; higher is better.
Metric: Pass@5 (%; any of five generations judged correct, hybrid dense+lexical retrieval). Source: arxiv.org. Saturation forecast: Around September 2027. 8 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Sonnet 4.6 | 65.1 | #158 |
| 2 | GPT-5.5 | 65 | #48 |
| 3 | Gemini 3.1 Pro (Preview) | 63.9 | #91 |
| 4 | Qwen 3 235B A22B | 63.7 | #622 |
| 5 | Qwen 3 30B A3B | 55.3 | #1124 |
| 6 | Qwen 3 8B | 46.1 | #1838 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=stage-cross-scene-reasoning-hybrid-rag · How It Works · Data refreshed daily, snapshot 2026-11-12.