CompliBench - Airline - Violation Detection: leaderboard
Metric: Violation detection accuracy (%): share of violating assistant turns in the airline domain that the judge flags as violations, of an LLM judge on CompliBench's synthesized multi-turn dialogues with injected, adversarially optimized guideline violations, mean of four runs at default reasoning effort; higher is better. Source: arxiv.org. Saturation forecast: Estimated already saturated. 15 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3 Pro | 90.61 |
| 2 | Kimi K2.5 | 86.88 |
| 3 | GLM-5 | 86.46 |
| 4 | DeepSeek V3.2 (Thinking) | 84.81 |
| 5 | Qwen 3.5 Plus | 82.32 |
| 6 | Claude Sonnet 4.6 | 81.76 |
| 7 | GPT-5 | 81.08 |
| 8 | Qwen 3 Max | 71.41 |
| 9 | Qwen 3 30B A3B | 69.75 |
| 10 | GPT-4o | 50.55 |
| 11 | Qwen 3 32B | 41.44 |
| 12 | Qwen 3 14B | 32.04 |
| 13 | Qwen 3 8B | 28.18 |
| 14 | Qwen 3 4B | 24.86 |
| 15 | GPT-4o Mini | 19.61 |
Interactive version: theaggregate.ai/benchmark?slug=complibench-airline-violation-detection · How It Works · Data refreshed daily, snapshot 2026-10-07.