MiGUE-Bench - Cross-Document Causal Relations: leaderboard
Metric: Accuracy (%) on 300 cross-document causal relation multiple-choice questions with LLM-built distractors over news documents; higher is better. Source: arxiv.org. Saturation forecast: Around March 2027. 12 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Claude Opus 4.5 | 70.48 |
| 2 | GPT-5.2 Pro | 64.08 |
| 3 | DeepSeek V3.2 | 62.86 |
| 4 | Qwen 3 Max | 60 |
| 5 | Gemini 3 Pro | 55.71 |
| 6 | GLM-4.7 | 53.33 |
| 7 | Qwen 3 235B A22B | 52.86 |
| 8 | Kimi K2 | 51.43 |
| 9 | Qwen 3 30B A3B | 50.95 |
| 10 | Claude Haiku 4.5 | 42.38 |
| 11 | Qwen 3 8B | 33.9 |
Interactive version: theaggregate.ai/benchmark?slug=migue-bench-cross-document-causal-relations · How It Works · Data refreshed daily, snapshot 2026-09-29.