LongJud-Bench - Fact Marshalling (Second Instance): leaderboard
Metric: Fact Marshalling Score (%; Claude Sonnet 4.6 judge, 0-10 rubric on the appeal reasons in the second-instance appeal or response, normalized). Source: arxiv.org. Saturation forecast: Around December 2026. 6 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Kimi K2.5 | 74 |
| 2 | Qwen 3.5 Flash | 72 |
| 3 | GLM-4.7 | 70 |
| 4 | Qwen 3.5 Plus | 70 |
| 5 | DeepSeek V4 Flash | 69 |
| 6 | GPT-5.2 | 68 |
Interactive version: theaggregate.ai/benchmark?slug=longjud-bench-fact-marshalling-second-instance · How It Works · Data refreshed daily, snapshot 2026-09-25.