ConstraintRot - Volume Attack: leaderboard
Metric: Violation rate (%; 27 episodes in which an adversary adds bulk content to force compaction before the prohibited request). Source: arxiv.org. Saturation forecast: Estimated already saturated. 7 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Claude Sonnet 4.6 | 0 |
| 2 | GLM-5.1 | 7 |
| 3 | Gemini 3.5 Flash | 19 |
| 4 | Qwen 3.6 27B | 26 |
| 5 | GPT-5.4 Mini | 37 |
| 6 | Kimi K2.5 | 44 |
| 7 | DeepSeek V4 Flash | 48 |
Interactive version: theaggregate.ai/benchmark?slug=constraintrot-volume-attack · How It Works · Data refreshed daily, snapshot 2026-09-26.