COMPL-AI: leaderboard
Up to 23 benchmarks mapped to technical requirements under six EU AI Act principles, each scored 0 to 1, from ETH Zurich, INSAIT and LatticeFlow AI (2024); 12 LLMs assessed in the paper.
Metric: COMPL-AI average (score). Source: arxiv.org. Source: benchmarklist.com. Saturation forecast: Estimated already saturated. 15 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-4 Preview (1106) | 0.86 | #418 |
| 2 | Claude 3 Opus | 0.85 | #571 |
| 3 | Gemini 1.5 Flash (001) | 0.8 | #577 |
| 4 | GPT-3.5 Turbo (0125) | 0.77 | #984 |
| 5 | Yi Chat 34B | 0.72 | #976 |
| 6 | Qwen 1.5 72B Chat | 0.72 | #776 |
| 7 | Bielik-11B-v2.3-Instruct | 0.71 | #693 |
| 8 | Llama 2 70B Chat (HF) | 0.7 | #1264 |
| 9 | Mixtral 8x7B Instruct (v0.1) | 0.7 | #1073 |
| 10 | Mistral 7B Instruct (v0.3) | 0.68 | #1274 |
| 11 | Mistral 7B Instruct (v0.2) | 0.67 | #1333 |
| 12 | Llama 2 13B Chat Base | 0.66 | #1422 |
| 13 | Mistral-7B-v0.3 | 0.66 | #1355 |
| 14 | Llama 2 7B Chat | 0.63 | #1499 |
| 15 | Gemma 2 9B | 0.58 | #958 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=compl-ai · How It Works · Data refreshed daily, snapshot 2026-10-20.