CompliBench - Healthcare - Violation Detection: leaderboard
CompliBench (2026) evaluates LLMs as judges of guideline compliance in task-oriented dialogue: synthesized multi-turn conversations carry injected, adversarially optimized violations, and the judge must name the governing guideline and violation status of every assistant turn. This board is violation detection accuracy in percent in the healthcare domain, the share of violating turns the judge flags; higher is better.
Metric: Violation detection accuracy (%): share of violating assistant turns in the healthcare 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. Updated 7 Oct 2026.
Top models
| # | Model | Setting | Score | Overall rank |
|---|---|---|---|---|
| 1 | Gemini 3 Pro | 94.8 | #145 | |
| 2 | Claude Sonnet 4.6 | 90.37 | #173 | |
| 3 | Kimi K2.5 | 89.45 | #266 | |
| 4 | GLM-5 | 88.15 | #269 | |
| 5 | DeepSeek V3.2 | Thinking | 82.49 | #419 |
| 6 | Qwen 3.5 Plus | 80.5 | #242 | |
| 7 | Qwen 3 30B A3B | 77.83 | #1156 | |
| 8 | GPT-5 | 77.68 | #178 | |
| 9 | Qwen 3 Max | 72.71 | #415 | |
| 10 | GPT-4o | 57.03 | #751 | |
| 11 | Qwen 3 32B | 43.43 | #972 | |
| 12 | Qwen 3 14B | 31.35 | #1269 | |
| 13 | Qwen 3 8B | 25.99 | #1873 | |
| 14 | Qwen 3 4B | 22.02 | #2348 | |
| 15 | GPT-4o Mini | 15.37 | #1667 |
No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.
Interactive version: theaggregate.ai/benchmark?slug=complibench-healthcare-violation-detection · How It Works · Data refreshed daily, snapshot 2026-11-21.