CompliBench - Insurance - Guideline Accuracy: 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 strict guideline accuracy in percent in the insurance domain, the share of compliant turns for which the judge names the right guideline and no violation; higher is better.

Metric: Strict guideline accuracy (%): share of compliant assistant turns in the insurance domain for which the judge predicts the correct governing guideline and no violation, 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

#ModelSettingScoreOverall rank
1GPT-591.91#178
2Gemini 3 Pro91.68#145
3Qwen 3.5 Plus90.8#242
4Qwen 3 Max89.82#415
5Claude Sonnet 4.685.71#173
6DeepSeek V3.2Thinking85.5#419
7GLM-584.62#269
8Qwen 3 30B A3B82.87#1156
9GPT-4o82.7#751
10Kimi K2.578.67#266
11Qwen 3 32B65.05#972
12GPT-4o Mini56.76#1667
13Qwen 3 4B51.01#2348
14Qwen 3 8B48.63#1873
15Qwen 3 14B41.97#1269

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=complibench-insurance-guideline-accuracy · How It Works · Data refreshed daily, snapshot 2026-11-21.